diff --git a/MODULE.bazel b/MODULE.bazel
index 3eaedcc6..4042e925 100644
--- a/MODULE.bazel
+++ b/MODULE.bazel
@@ -126,6 +126,13 @@ http_archive(
     urls = ["https://maven.pkg.github.com/NVIDIA/ncore/com/nvidia/ncore/test-data-v4/0.2/test-data-v4-0.2.tar.gz"],
 )
 
+# time-trimmed zarr format v2 / v3 serializations of the test-data-v4 sequence
+http_archive(
+    name = "test-data-v4-trimmed",
+    sha256 = "a4a3457829431e30cdc5bb55cc00a8e6d5dfdc5d52db2221594807b4c9fff81d",
+    urls = ["https://maven.pkg.github.com/NVIDIA/ncore/com/nvidia/ncore/test-data-v4-trimmed/0.2/test-data-v4-trimmed-0.2.tar.gz"],
+)
+
 # documentation binary data
 http_archive(
     name = "ncore_docs_data",
diff --git a/MODULE.bazel.lock b/MODULE.bazel.lock
index 7d07a1da..63996be3 100644
--- a/MODULE.bazel.lock
+++ b/MODULE.bazel.lock
@@ -1502,9 +1502,9 @@
         "bzlTransitiveDigest": "zc0QWNGxYYxPvHvT1CA/9+X0h1JuhXagw7k7q1jOhwg=",
         "usagesDigest": "rf/YOtMUZwRHGFAE+9knWdjNz4HOysfZIlX5MJbhT+E=",
         "recordedFileInputs": {
-          "@@//deps/pip/requirements_3_11.txt": "845ddefb3a79c099a05f39880752241d3c30cefb3774c0611f253d5bff3f774f",
+          "@@//deps/pip/requirements_3_11.txt": "a13d4223fd7b56af9192342a5af35245f7678050a5d3e33fd5ea686da8e79680",
           "@@//deps/pip/requirements_3_12.txt": "56ddea672a89523c8b93e54bcbcffc28c91aeff4f0e72c8b95306dc287af0068",
-          "@@//deps/pip/requirements_3_13.txt": "48cd877f84a1ac6277b6cd9d8f5c211f2f2b49e058dc1f1c0717e3b1bfbfa689",
+          "@@//deps/pip/requirements_3_13.txt": "81e3dc5cc6fd0ad9c8a3245c6645328adf2bbc40040f84f10b3cb9b43278e9fb",
           "@@//deps/pip/requirements_3_8.txt": "94c254d645dc1d0555f48e66fa499aa2efecfc5e547500cd4ed409aee28e8880",
           "@@rules_fuzzing+//fuzzing/requirements.txt": "ab04664be026b632a0d2a2446c4f65982b7654f5b6851d2f9d399a19b7242a5b",
           "@@rules_python+//tools/publish/requirements_darwin.txt": "095d4a4f3d639dce831cd493367631cd51b53665292ab20194bac2c0c6458fa8",
@@ -1733,36 +1733,6 @@
               ]
             }
           },
-          "ncore_pip_deps_311_asciitree_sdist_4aa4b9b6": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp311_linux_aarch64",
-                "cp311_linux_arm",
-                "cp311_linux_ppc",
-                "cp311_linux_s390x",
-                "cp311_linux_x86_64",
-                "cp311_osx_aarch64",
-                "cp311_osx_x86_64",
-                "cp311_windows_x86_64"
-              ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cu128/torch"
-              ],
-              "filename": "asciitree-0.3.3.tar.gz",
-              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
-              "repo": "ncore_pip_deps_311",
-              "requirement": "asciitree==0.3.3",
-              "sha256": "4aa4b9b649f85e3fcb343363d97564aa1fb62e249677f2e18a96765145cc0f6e",
-              "urls": [
-                "https://files.pythonhosted.org/packages/2d/6a/885bc91484e1aa8f618f6f0228d76d0e67000b0fdd6090673b777e311913/asciitree-0.3.3.tar.gz"
-              ]
-            }
-          },
           "ncore_pip_deps_311_asttokens_py3_none_any_15a3ebc0": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
@@ -3545,6 +3515,60 @@
               ]
             }
           },
+          "ncore_pip_deps_311_donfig_py3_none_any_2a3175ce": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "donfig-0.8.1.post1-py3-none-any.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "donfig==0.8.1.post1",
+              "sha256": "2a3175ce74a06109ff9307d90a230f81215cbac9a751f4d1c6194644b8204f9d",
+              "urls": [
+                "https://files.pythonhosted.org/packages/0c/d5/c5db1ea3394c6e1732fb3286b3bd878b59507a8f77d32a2cebda7d7b7cd4/donfig-0.8.1.post1-py3-none-any.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_donfig_sdist_3bef3413": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cu128/torch"
+              ],
+              "filename": "donfig-0.8.1.post1.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "donfig==0.8.1.post1",
+              "sha256": "3bef3413a4c1c601b585e8d297256d0c1470ea012afa6e8461dc28bfb7c23f52",
+              "urls": [
+                "https://files.pythonhosted.org/packages/25/71/80cc718ff6d7abfbabacb1f57aaa42e9c1552bfdd01e64ddd704e4a03638/donfig-0.8.1.post1.tar.gz"
+              ]
+            }
+          },
           "ncore_pip_deps_311_dracopy_cp311_cp311_macosx_10_9_x86_64_c4131151": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
@@ -3791,60 +3815,6 @@
               ]
             }
           },
-          "ncore_pip_deps_311_fasteners_py3_none_any_758819cb": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp311_linux_aarch64",
-                "cp311_linux_arm",
-                "cp311_linux_ppc",
-                "cp311_linux_s390x",
-                "cp311_linux_x86_64",
-                "cp311_osx_aarch64",
-                "cp311_osx_x86_64",
-                "cp311_windows_x86_64"
-              ],
-              "filename": "fasteners-0.19-py3-none-any.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
-              "repo": "ncore_pip_deps_311",
-              "requirement": "fasteners==0.19",
-              "sha256": "758819cb5d94cdedf4e836988b74de396ceacb8e2794d21f82d131fd9ee77237",
-              "urls": [
-                "https://files.pythonhosted.org/packages/61/bf/fd60001b3abc5222d8eaa4a204cd8c0ae78e75adc688f33ce4bf25b7fafa/fasteners-0.19-py3-none-any.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_311_fasteners_sdist_b4f37c3a": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp311_linux_aarch64",
-                "cp311_linux_arm",
-                "cp311_linux_ppc",
-                "cp311_linux_s390x",
-                "cp311_linux_x86_64",
-                "cp311_osx_aarch64",
-                "cp311_osx_x86_64",
-                "cp311_windows_x86_64"
-              ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cu128/torch"
-              ],
-              "filename": "fasteners-0.19.tar.gz",
-              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
-              "repo": "ncore_pip_deps_311",
-              "requirement": "fasteners==0.19",
-              "sha256": "b4f37c3ac52d8a445af3a66bce57b33b5e90b97c696b7b984f530cf8f0ded09c",
-              "urls": [
-                "https://files.pythonhosted.org/packages/5f/d4/e834d929be54bfadb1f3e3b931c38e956aaa3b235a46a3c764c26c774902/fasteners-0.19.tar.gz"
-              ]
-            }
-          },
           "ncore_pip_deps_311_fastjsonschema_py3_none_any_c9e5b7e9": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
@@ -4361,6 +4331,156 @@
               ]
             }
           },
+          "ncore_pip_deps_311_google_crc32c_cp311_cp311_macosx_12_0_arm64_457d0d9a": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp311-cp311-macosx_12_0_arm64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "457d0d9a4718fd52b1494eac5c200ad25beeadbdc91843d550a003910838589f",
+              "urls": [
+                "https://files.pythonhosted.org/packages/0e/55/a2f07f15e624f0de79359b1a6c1deb59ec5061bd3b38744b3b2849400662/google_crc32c-1.9.0-cp311-cp311-macosx_12_0_arm64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_google_crc32c_cp311_cp311_macosx_12_0_x86_64_ccfe4002": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp311-cp311-macosx_12_0_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "ccfe40021fd6afe23361175cf7551e3cef5fd34dc1ebe319f14993a83579e0eb",
+              "urls": [
+                "https://files.pythonhosted.org/packages/f8/b3/923743597b774bbcf12a7c3e00e48d745e15fd616ad7489a40a63fff8f2f/google_crc32c-1.9.0-cp311-cp311-macosx_12_0_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_google_crc32c_cp311_cp311_manylinux_2_17_aarch64_fbef61a3": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "fbef61a3794e011c65fb4396a196cf123a7f474fe5a443db8e5dd7d751b9e6d4",
+              "urls": [
+                "https://files.pythonhosted.org/packages/df/a6/4d0352fe889663e0d81cea7fc664ec9158727384de4a44ab10e9967a7682/google_crc32c-1.9.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_google_crc32c_cp311_cp311_manylinux_2_17_x86_64_86764b99": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "86764b99e7a607830d93cb5b75e0ec3ff6cb06d3c274624418473cee701900d4",
+              "urls": [
+                "https://files.pythonhosted.org/packages/aa/e3/26685384e4b66ff0928d9566ef6110a7df76029175a1842329d7e3515f10/google_crc32c-1.9.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_google_crc32c_cp311_cp311_win_amd64_43a2dc26": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp311-cp311-win_amd64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "43a2dc26f9be213fbe0b4fc4a1088c5d45cbfcb3247420ccc820f0fc3edeea86",
+              "urls": [
+                "https://files.pythonhosted.org/packages/cb/ce/4e90102e84880e97d3cf935f2672ecd29191bdeacf57f01740f92debda00/google_crc32c-1.9.0-cp311-cp311-win_amd64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_google_crc32c_sdist_7b8c84c3": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cu128/torch"
+              ],
+              "filename": "google_crc32c-1.9.0.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "7b8c84c3d159ab6817fe3f74e6e6cef099c3f95dcec3abc0d8afb1404642efbe",
+              "urls": [
+                "https://files.pythonhosted.org/packages/fa/25/9cb0c1c31c45b893eb8f11ae70b3f4309432d59b5acaebca5dbe791729a4/google_crc32c-1.9.0.tar.gz"
+              ]
+            }
+          },
           "ncore_pip_deps_311_google_pasta_py3_none_any_b3248279": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
@@ -12269,7 +12389,7 @@
               ]
             }
           },
-          "ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_15_universal2_c0598838": {
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_macosx_10_13_x86_64_44edc647": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -12283,17 +12403,17 @@
                 "cp311_osx_x86_64",
                 "cp311_windows_x86_64"
               ],
-              "filename": "pyzmq-26.2.1-cp311-cp311-macosx_10_15_universal2.whl",
+              "filename": "pyyaml-6.0.3-cp311-cp311-macosx_10_13_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "pyzmq==26.2.1",
-              "sha256": "c059883840e634a21c5b31d9b9a0e2b48f991b94d60a811092bc37992715146a",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e",
               "urls": [
-                "https://files.pythonhosted.org/packages/b9/03/5ecc46a6ed5971299f5c03e016ca637802d8660e44392bea774fb7797405/pyzmq-26.2.1-cp311-cp311-macosx_10_15_universal2.whl"
+                "https://files.pythonhosted.org/packages/6d/16/a95b6757765b7b031c9374925bb718d55e0a9ba8a1b6a12d25962ea44347/pyyaml-6.0.3-cp311-cp311-macosx_10_13_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_9_x86_64_ed038a92": {
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_macosx_11_0_arm64_652cb6ed": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -12307,17 +12427,17 @@
                 "cp311_osx_x86_64",
                 "cp311_windows_x86_64"
               ],
-              "filename": "pyzmq-26.2.1-cp311-cp311-macosx_10_9_x86_64.whl",
+              "filename": "pyyaml-6.0.3-cp311-cp311-macosx_11_0_arm64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "pyzmq==26.2.1",
-              "sha256": "ed038a921df836d2f538e509a59cb638df3e70ca0fcd70d0bf389dfcdf784d2a",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824",
               "urls": [
-                "https://files.pythonhosted.org/packages/40/51/48fec8f990ee644f461ff14c8fe5caa341b0b9b3a0ad7544f8ef17d6f528/pyzmq-26.2.1-cp311-cp311-macosx_10_9_x86_64.whl"
+                "https://files.pythonhosted.org/packages/16/19/13de8e4377ed53079ee996e1ab0a9c33ec2faf808a4647b7b4c0d46dd239/pyyaml-6.0.3-cp311-cp311-macosx_11_0_arm64.whl"
               ]
             }
           },
-          "ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_aarch64_9027a7fc": {
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_aarch64_10892704": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -12331,17 +12451,17 @@
                 "cp311_osx_x86_64",
                 "cp311_windows_x86_64"
               ],
-              "filename": "pyzmq-26.2.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
+              "filename": "pyyaml-6.0.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "pyzmq==26.2.1",
-              "sha256": "9027a7fcf690f1a3635dc9e55e38a0d6602dbbc0548935d08d46d2e7ec91f454",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c",
               "urls": [
-                "https://files.pythonhosted.org/packages/c1/f4/f322b389727c687845e38470b48d7a43c18a83f26d4d5084603c6c3f79ca/pyzmq-26.2.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl"
+                "https://files.pythonhosted.org/packages/0c/62/d2eb46264d4b157dae1275b573017abec435397aa59cbcdab6fc978a8af4/pyyaml-6.0.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_x86_64_f0019cc8": {
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_s390x_850774a7": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -12355,11 +12475,233 @@
                 "cp311_osx_x86_64",
                 "cp311_windows_x86_64"
               ],
-              "filename": "pyzmq-26.2.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl",
+              "filename": "pyyaml-6.0.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "pyzmq==26.2.1",
-              "sha256": "f0019cc804ac667fb8c8eaecdb66e6d4a68acf2e155d5c7d6381a5645bd93ae4",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00",
+              "urls": [
+                "https://files.pythonhosted.org/packages/10/cb/16c3f2cf3266edd25aaa00d6c4350381c8b012ed6f5276675b9eba8d9ff4/pyyaml-6.0.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_x86_64_b8bb0864": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d",
+              "urls": [
+                "https://files.pythonhosted.org/packages/71/60/917329f640924b18ff085ab889a11c763e0b573da888e8404ff486657602/pyyaml-6.0.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_musllinux_1_2_aarch64_1d37d57a": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp311-cp311-musllinux_1_2_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a",
+              "urls": [
+                "https://files.pythonhosted.org/packages/dd/6f/529b0f316a9fd167281a6c3826b5583e6192dba792dd55e3203d3f8e655a/pyyaml-6.0.3-cp311-cp311-musllinux_1_2_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_musllinux_1_2_x86_64_37503bfb": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp311-cp311-musllinux_1_2_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4",
+              "urls": [
+                "https://files.pythonhosted.org/packages/f2/6a/b627b4e0c1dd03718543519ffb2f1deea4a1e6d42fbab8021936a4d22589/pyyaml-6.0.3-cp311-cp311-musllinux_1_2_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyyaml_cp311_cp311_win_amd64_9f3bfb49": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp311-cp311-win_amd64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf",
+              "urls": [
+                "https://files.pythonhosted.org/packages/da/e3/ea007450a105ae919a72393cb06f122f288ef60bba2dc64b26e2646fa315/pyyaml-6.0.3-cp311-cp311-win_amd64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyyaml_sdist_d7662337": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cu128/torch"
+              ],
+              "filename": "pyyaml-6.0.3.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f",
+              "urls": [
+                "https://files.pythonhosted.org/packages/05/8e/961c0007c59b8dd7729d542c61a4d537767a59645b82a0b521206e1e25c2/pyyaml-6.0.3.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_15_universal2_c0598838": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyzmq-26.2.1-cp311-cp311-macosx_10_15_universal2.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyzmq==26.2.1",
+              "sha256": "c059883840e634a21c5b31d9b9a0e2b48f991b94d60a811092bc37992715146a",
+              "urls": [
+                "https://files.pythonhosted.org/packages/b9/03/5ecc46a6ed5971299f5c03e016ca637802d8660e44392bea774fb7797405/pyzmq-26.2.1-cp311-cp311-macosx_10_15_universal2.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_9_x86_64_ed038a92": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyzmq-26.2.1-cp311-cp311-macosx_10_9_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyzmq==26.2.1",
+              "sha256": "ed038a921df836d2f538e509a59cb638df3e70ca0fcd70d0bf389dfcdf784d2a",
+              "urls": [
+                "https://files.pythonhosted.org/packages/40/51/48fec8f990ee644f461ff14c8fe5caa341b0b9b3a0ad7544f8ef17d6f528/pyzmq-26.2.1-cp311-cp311-macosx_10_9_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_aarch64_9027a7fc": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyzmq-26.2.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyzmq==26.2.1",
+              "sha256": "9027a7fcf690f1a3635dc9e55e38a0d6602dbbc0548935d08d46d2e7ec91f454",
+              "urls": [
+                "https://files.pythonhosted.org/packages/c1/f4/f322b389727c687845e38470b48d7a43c18a83f26d4d5084603c6c3f79ca/pyzmq-26.2.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_x86_64_f0019cc8": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp311_linux_aarch64",
+                "cp311_linux_arm",
+                "cp311_linux_ppc",
+                "cp311_linux_s390x",
+                "cp311_linux_x86_64",
+                "cp311_osx_aarch64",
+                "cp311_osx_x86_64",
+                "cp311_windows_x86_64"
+              ],
+              "filename": "pyzmq-26.2.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
+              "repo": "ncore_pip_deps_311",
+              "requirement": "pyzmq==26.2.1",
+              "sha256": "f0019cc804ac667fb8c8eaecdb66e6d4a68acf2e155d5c7d6381a5645bd93ae4",
               "urls": [
                 "https://files.pythonhosted.org/packages/b5/e2/45c0f6e122b562cb8c6c45c0dcac1160a4e2207385ef9b13463e74f93031/pyzmq-26.2.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"
               ]
@@ -16919,7 +17261,7 @@
               ]
             }
           },
-          "ncore_pip_deps_311_zarr_py3_none_any_2795e20a": {
+          "ncore_pip_deps_311_zarr_py3_none_any_72c3dcee": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -16933,17 +17275,17 @@
                 "cp311_osx_x86_64",
                 "cp311_windows_x86_64"
               ],
-              "filename": "zarr-2.18.4-py3-none-any.whl",
+              "filename": "zarr-3.1.4-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "zarr==2.18.4",
-              "sha256": "2795e20aff91093ce7e4da36ab1a138aededbd8ab66bf01fd01512e61d31e5d1",
+              "requirement": "zarr==3.1.4",
+              "sha256": "72c3dcee2d41527d21cd814cbaca1e6e23550701eb4fc30767ee53d925900eec",
               "urls": [
-                "https://files.pythonhosted.org/packages/b4/d1/c84022a44afc7b7ccc442fba3daee56bdd03593d91ee4bc245a08e4fcc55/zarr-2.18.4-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/2d/8e/1b05f5b8a9fb1db01d1c09c3020125fc5c278d97843233dd6fa897a5f24f/zarr-3.1.4-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_311_zarr_sdist_37790ede": {
+          "ncore_pip_deps_311_zarr_sdist_b0cf4eea": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -16963,13 +17305,13 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cu128/torch"
               ],
-              "filename": "zarr-2.18.4.tar.gz",
+              "filename": "zarr-3.1.4.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_11_host//:python",
               "repo": "ncore_pip_deps_311",
-              "requirement": "zarr==2.18.4",
-              "sha256": "37790ededd0683ae1abe6ff90aa16c22543b3436810060f53d72c15e910c24bb",
+              "requirement": "zarr==3.1.4",
+              "sha256": "b0cf4eea8367ae444d7d0ea90b98d29ac0b7eb691fac2e0bcc2d2bd659522903",
               "urls": [
-                "https://files.pythonhosted.org/packages/21/d1/764ca5b66d91b20dede66aedc6eb9ede3adbe5c61779e7378a7ecb010e87/zarr-2.18.4.tar.gz"
+                "https://files.pythonhosted.org/packages/c6/77/170915815b797861e0d74eb8246da2898f4069838d5f638de5d8bd1ab7c7/zarr-3.1.4.tar.gz"
               ]
             }
           },
@@ -20309,36 +20651,6 @@
               ]
             }
           },
-          "ncore_pip_deps_313_asciitree_sdist_4aa4b9b6": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "asciitree-0.3.3.tar.gz",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "asciitree==0.3.3",
-              "sha256": "4aa4b9b649f85e3fcb343363d97564aa1fb62e249677f2e18a96765145cc0f6e",
-              "urls": [
-                "https://files.pythonhosted.org/packages/2d/6a/885bc91484e1aa8f618f6f0228d76d0e67000b0fdd6090673b777e311913/asciitree-0.3.3.tar.gz"
-              ]
-            }
-          },
           "ncore_pip_deps_313_cbor2_cp313_cp313_macosx_11_0_arm64_f850860e": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
@@ -20567,7 +20879,7 @@
               ]
             }
           },
-          "ncore_pip_deps_313_deprecated_py3_none_any_58204cf4": {
+          "ncore_pip_deps_313_donfig_py3_none_any_2a3175ce": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20581,17 +20893,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "deprecated-3.0.0-py3-none-any.whl",
+              "filename": "donfig-0.8.1.post1-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "deprecated==3.0.0",
-              "sha256": "58204cf4a7f6270d547af5c278ee7a6bb56045a4b3d8441a1cd11660f41b7939",
+              "requirement": "donfig==0.8.1.post1",
+              "sha256": "2a3175ce74a06109ff9307d90a230f81215cbac9a751f4d1c6194644b8204f9d",
               "urls": [
-                "https://files.pythonhosted.org/packages/83/ae/676feae8e4644a6d7169951a97f61c56f416c73f67bf1761f2461d75cc81/deprecated-3.0.0-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/0c/d5/c5db1ea3394c6e1732fb3286b3bd878b59507a8f77d32a2cebda7d7b7cd4/donfig-0.8.1.post1-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_deprecated_sdist_16850204": {
+          "ncore_pip_deps_313_donfig_sdist_3bef3413": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20611,17 +20923,17 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "deprecated-3.0.0.tar.gz",
+              "filename": "donfig-0.8.1.post1.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "deprecated==3.0.0",
-              "sha256": "16850204d3a1e6bb0acd06bff48d96e8b0a0d25d1c52f71705405a0f4894192d",
+              "requirement": "donfig==0.8.1.post1",
+              "sha256": "3bef3413a4c1c601b585e8d297256d0c1470ea012afa6e8461dc28bfb7c23f52",
               "urls": [
-                "https://files.pythonhosted.org/packages/f7/9c/16649913bf14c73e0a9453782e148362ff2657067deff6aa9c7ebcddcc31/deprecated-3.0.0.tar.gz"
+                "https://files.pythonhosted.org/packages/25/71/80cc718ff6d7abfbabacb1f57aaa42e9c1552bfdd01e64ddd704e4a03638/donfig-0.8.1.post1.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_fasteners_py3_none_any_9422c40d": {
+          "ncore_pip_deps_313_filelock_py3_none_any_2ce9818e": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20635,17 +20947,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "fasteners-0.20-py3-none-any.whl",
+              "filename": "filelock-4.1.0-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "fasteners==0.20",
-              "sha256": "9422c40d1e350e4259f509fb2e608d6bc43c0136f79a00db1b49046029d0b3b7",
+              "requirement": "filelock==4.1.0",
+              "sha256": "2ce9818e3e2d8f284c1a964414447ef148d42a5fd5e2a477a7118e574b293ec1",
               "urls": [
-                "https://files.pythonhosted.org/packages/51/ac/e5d886f892666d2d1e5cb8c1a41146e1d79ae8896477b1153a21711d3b44/fasteners-0.20-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/ee/86/032133892a5de43b5a98200b01aadcad68cc255e274a762f08b8a76d2912/filelock-4.1.0-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_fasteners_sdist_55dce879": {
+          "ncore_pip_deps_313_filelock_sdist_ad7f724a": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20665,17 +20977,17 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "fasteners-0.20.tar.gz",
+              "filename": "filelock-4.1.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "fasteners==0.20",
-              "sha256": "55dce8792a41b56f727ba6e123fcaee77fd87e638a6863cec00007bfea84c8d8",
+              "requirement": "filelock==4.1.0",
+              "sha256": "ad7f724afef953e731b1cc39bcd3a09166d72ed7fcdf29e6e88b1c3235c6715d",
               "urls": [
-                "https://files.pythonhosted.org/packages/2d/18/7881a99ba5244bfc82f06017316ffe93217dbbbcfa52b887caa1d4f2a6d3/fasteners-0.20.tar.gz"
+                "https://files.pythonhosted.org/packages/4c/58/6fd434bec86eff7c38a3168454cb132b762b2bea9b3ac094101a2f7bc32a/filelock-4.1.0.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_filelock_py3_none_any_2ce9818e": {
+          "ncore_pip_deps_313_fsspec_py3_none_any_8dd6e646": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20689,17 +21001,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "filelock-4.1.0-py3-none-any.whl",
+              "filename": "fsspec-2026.9.0-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "filelock==4.1.0",
-              "sha256": "2ce9818e3e2d8f284c1a964414447ef148d42a5fd5e2a477a7118e574b293ec1",
+              "requirement": "fsspec==2026.9.0",
+              "sha256": "8dd6e646e99ea382bd85f97a45e6b526a442d79423a7dc673f1e2756d05fcb5f",
               "urls": [
-                "https://files.pythonhosted.org/packages/ee/86/032133892a5de43b5a98200b01aadcad68cc255e274a762f08b8a76d2912/filelock-4.1.0-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/6c/c0/a98505f18594f1bce828bb159cec0fcf9860562f1a2c85913409fc8f3d9e/fsspec-2026.9.0-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_filelock_sdist_ad7f724a": {
+          "ncore_pip_deps_313_fsspec_sdist_0f081479": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20719,17 +21031,17 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "filelock-4.1.0.tar.gz",
+              "filename": "fsspec-2026.9.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "filelock==4.1.0",
-              "sha256": "ad7f724afef953e731b1cc39bcd3a09166d72ed7fcdf29e6e88b1c3235c6715d",
+              "requirement": "fsspec==2026.9.0",
+              "sha256": "0f08147951c8cb31d844c3547d631053b127863b60be04cf06e121333ee0e2fe",
               "urls": [
-                "https://files.pythonhosted.org/packages/4c/58/6fd434bec86eff7c38a3168454cb132b762b2bea9b3ac094101a2f7bc32a/filelock-4.1.0.tar.gz"
+                "https://files.pythonhosted.org/packages/77/cd/9be253869fc42e764de7f3dedd6969af7d44ff9c3375214a3442a6f3fc08/fsspec-2026.9.0.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_fsspec_py3_none_any_8dd6e646": {
+          "ncore_pip_deps_313_google_crc32c_cp313_cp313_macosx_12_0_arm64_f1dc17d9": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20743,17 +21055,113 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "fsspec-2026.9.0-py3-none-any.whl",
+              "filename": "google_crc32c-1.9.0-cp313-cp313-macosx_12_0_arm64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "fsspec==2026.9.0",
-              "sha256": "8dd6e646e99ea382bd85f97a45e6b526a442d79423a7dc673f1e2756d05fcb5f",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "f1dc17d987ddcc5eba12a7ce48f0eb93141dea236b170c1101151396edf2f0cf",
               "urls": [
-                "https://files.pythonhosted.org/packages/6c/c0/a98505f18594f1bce828bb159cec0fcf9860562f1a2c85913409fc8f3d9e/fsspec-2026.9.0-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/3f/34/cb484e8b6174f130f8c6dc79c733a9dd8869b410ad6511fb6104c46b973a/google_crc32c-1.9.0-cp313-cp313-macosx_12_0_arm64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_fsspec_sdist_0f081479": {
+          "ncore_pip_deps_313_google_crc32c_cp313_cp313_macosx_12_0_x86_64_f894a287": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp313-cp313-macosx_12_0_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "f894a2877650b56201d26a012a257b76d54a68834dc3913a93830ca8a047b075",
+              "urls": [
+                "https://files.pythonhosted.org/packages/af/25/3e8e567bd48448e225ea27318ccf2b94e05124e7b8b97b13eaec9e127199/google_crc32c-1.9.0-cp313-cp313-macosx_12_0_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_google_crc32c_cp313_cp313_manylinux_2_17_aarch64_4488f155": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "4488f1553a9ab7e86cdedc833374a7e904031803b995dc0bd0be48c271fa6556",
+              "urls": [
+                "https://files.pythonhosted.org/packages/f0/18/bee0dd59ae622482dc6463636c79e4bde7c954d061c859c9256362c9931a/google_crc32c-1.9.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_google_crc32c_cp313_cp313_manylinux_2_17_x86_64_0568b17e": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "0568b17ed90ac596f29400d99e243fd0cc6276766183def888d1bf8d1dc13827",
+              "urls": [
+                "https://files.pythonhosted.org/packages/fd/b6/e76e80fed5f2558273c7839e622f98095c9b36c719c7147e38e3c055cb70/google_crc32c-1.9.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_google_crc32c_cp313_cp313_win_amd64_8583ec21": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "google_crc32c-1.9.0-cp313-cp313-win_amd64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "8583ec21d56b565d68ab2963cc7e21b3b271247c29b04286068255ef65f221bd",
+              "urls": [
+                "https://files.pythonhosted.org/packages/87/34/165542bfa99dfef91a76471cc48cce74b8ff4e295722896087ab2b8e8611/google_crc32c-1.9.0-cp313-cp313-win_amd64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_google_crc32c_sdist_7b8c84c3": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -20773,13 +21181,13 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "fsspec-2026.9.0.tar.gz",
+              "filename": "google_crc32c-1.9.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "fsspec==2026.9.0",
-              "sha256": "0f08147951c8cb31d844c3547d631053b127863b60be04cf06e121333ee0e2fe",
+              "requirement": "google-crc32c==1.9.0",
+              "sha256": "7b8c84c3d159ab6817fe3f74e6e6cef099c3f95dcec3abc0d8afb1404642efbe",
               "urls": [
-                "https://files.pythonhosted.org/packages/77/cd/9be253869fc42e764de7f3dedd6969af7d44ff9c3375214a3442a6f3fc08/fsspec-2026.9.0.tar.gz"
+                "https://files.pythonhosted.org/packages/fa/25/9cb0c1c31c45b893eb8f11ae70b3f4309432d59b5acaebca5dbe791729a4/google_crc32c-1.9.0.tar.gz"
               ]
             }
           },
@@ -21293,31 +21701,7 @@
               ]
             }
           },
-          "ncore_pip_deps_313_mypy_extensions_py3_none_any_1be4cccd": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "mypy_extensions-1.1.0-py3-none-any.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "mypy-extensions==1.1.0",
-              "sha256": "1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505",
-              "urls": [
-                "https://files.pythonhosted.org/packages/79/7b/2c79738432f5c924bef5071f933bcc9efd0473bac3b4aa584a6f7c1c8df8/mypy_extensions-1.1.0-py3-none-any.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_mypy_extensions_sdist_52e68efc": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_macosx_10_13_x86_64_f13c127a": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21331,23 +21715,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "mypy_extensions-1.1.0.tar.gz",
+              "filename": "msgspec-0.22.0-cp313-cp313-macosx_10_13_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "mypy-extensions==1.1.0",
-              "sha256": "52e68efc3284861e772bbcd66823fde5ae21fd2fdb51c62a211403730b916558",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "f13c127a945479bc9db057eb253b8851075c8e1ae07ffc967bfa1c5676203a86",
               "urls": [
-                "https://files.pythonhosted.org/packages/a2/6e/371856a3fb9d31ca8dac321cda606860fa4548858c0cc45d9d1d4ca2628b/mypy_extensions-1.1.0.tar.gz"
+                "https://files.pythonhosted.org/packages/7f/62/5374fba2ede0408f4bd8b9b3a6c8464f8d0ea7ae9a2a064bd81ca492bd1e/msgspec-0.22.0-cp313-cp313-macosx_10_13_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_networkx_py3_none_any_e3fd2c13": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_macosx_11_0_arm64_5aa24eb4": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21361,17 +21739,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "networkx-3.7-py3-none-any.whl",
+              "filename": "msgspec-0.22.0-cp313-cp313-macosx_11_0_arm64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "networkx==3.7",
-              "sha256": "e3fd2c13a7814cee3746340d8d7f8598a67f16a58bf47fb7f8793fab6efca1b0",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "5aa24eb475d070ecbbe5b21080fc3ce4b0b76c60de25cfe0c9678d8fb44bb42f",
               "urls": [
-                "https://files.pythonhosted.org/packages/7e/cd/fe58041e9011f307c490e3e17dd48cc516448f7c698a3f2d9d9d65d7e6a8/networkx-3.7-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/cc/e3/357baa8d2a9164a98dfd7ef9d3a58125df0ed981be909945bdd337be7194/msgspec-0.22.0-cp313-cp313-macosx_11_0_arm64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_networkx_sdist_fd77a511": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_manylinux_2_17_aarch64_627bfdfe": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21385,23 +21763,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "networkx-3.7.tar.gz",
+              "filename": "msgspec-0.22.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "networkx==3.7",
-              "sha256": "fd77a511bd90f39f3d016351345b52cf5319b813bdca01de3f755d3cca62e96a",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "627bfdfe5a4b3d916b3360b30f4cddeee3a084f56593e33527c6872fa8322ff9",
               "urls": [
-                "https://files.pythonhosted.org/packages/dc/76/3af777226b63a5e64a6b36b1ec5855c14e2b94a37096d4760e595fc43511/networkx-3.7.tar.gz"
+                "https://files.pythonhosted.org/packages/fa/1b/9cc07718d1dee8ed5e89a265801d565bc0f15ead435ccb198f9c7bf92574/msgspec-0.22.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_10_13_x86_64_e3d82b70": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_manylinux_2_17_x86_64_c6c310ef": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21415,17 +21787,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numcodecs-0.15.1-cp313-cp313-macosx_10_13_x86_64.whl",
+              "filename": "msgspec-0.22.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numcodecs==0.15.1",
-              "sha256": "e3d82b70500cf61e8d115faa0d0a76be6ecdc24a16477ee3279d711699ad85f3",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "c6c310ef83e7e291b01a63298828f848348bb99e84a1098c4b3923c05674d032",
               "urls": [
-                "https://files.pythonhosted.org/packages/78/57/acbc54b3419e5be65015e47177c76c0a73e037fd3ae2cde5808169194d4d/numcodecs-0.15.1-cp313-cp313-macosx_10_13_x86_64.whl"
+                "https://files.pythonhosted.org/packages/46/64/f33fdfe95aca76601194a7064d14816c7c22c4eccc1b03a5335785895fa3/msgspec-0.22.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_11_0_arm64_1d471a18": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_musllinux_1_2_aarch64_bc374ded": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21439,17 +21811,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numcodecs-0.15.1-cp313-cp313-macosx_11_0_arm64.whl",
+              "filename": "msgspec-0.22.0-cp313-cp313-musllinux_1_2_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numcodecs==0.15.1",
-              "sha256": "1d471a1829ce52d3f365053a2bd1379e32e369517557c4027ddf5ac0d99c591e",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "bc374dedd5f85a5f4de2386dc5f737894ccb8c1ac18e9566ce66fd9839e6285d",
               "urls": [
-                "https://files.pythonhosted.org/packages/b6/56/9863fa6dc679f40a31bea5e9713ee5507a31dcd3ee82ea4b1a9268ce52e8/numcodecs-0.15.1-cp313-cp313-macosx_11_0_arm64.whl"
+                "https://files.pythonhosted.org/packages/88/a6/7b5c4fb39e0bf2dabc8be923c33c39b07ba769a0ce6f0afbbdfaadb1f2f2/msgspec-0.22.0-cp313-cp313-musllinux_1_2_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_x86_64_1dfdea4a": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_musllinux_1_2_x86_64_6f48317f": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21463,17 +21835,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numcodecs-0.15.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl",
+              "filename": "msgspec-0.22.0-cp313-cp313-musllinux_1_2_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numcodecs==0.15.1",
-              "sha256": "1dfdea4a67108205edfce99c1cb6cd621343bc7abb7e16a041c966776920e7de",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "6f48317f05312bfdf78248f53933f830f07ab75cc1c813ac3ca4220cb3b5b019",
               "urls": [
-                "https://files.pythonhosted.org/packages/fa/91/d96999b41e3146b6c0ce6bddc5ad85803cb4d743c95394562c2a4bb8cded/numcodecs-0.15.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"
+                "https://files.pythonhosted.org/packages/6c/e5/b4c5323b17ecfce45350695d40fc93e16856db957a53cbcf2f53007d6e12/msgspec-0.22.0-cp313-cp313-musllinux_1_2_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numcodecs_cp313_cp313_win_amd64_a4f7bdb2": {
+          "ncore_pip_deps_313_msgspec_cp313_cp313_win_amd64_0739b068": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21487,17 +21859,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numcodecs-0.15.1-cp313-cp313-win_amd64.whl",
+              "filename": "msgspec-0.22.0-cp313-cp313-win_amd64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numcodecs==0.15.1",
-              "sha256": "a4f7bdb26f1b34423cb56d48e75821223be38040907c9b5954eeb7463e7eb03c",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "0739b068f31f2004a364f97679ba91f2f5ecd6ec2a5b4b890188ab5c57d20672",
               "urls": [
-                "https://files.pythonhosted.org/packages/c3/32/233e5ede6568bdb044e6f99aaa9fa39827ff3109c6487fc137315f733586/numcodecs-0.15.1-cp313-cp313-win_amd64.whl"
+                "https://files.pythonhosted.org/packages/01/33/e591f9d3d8d6c9cfc02ae95f3e3c44920f2d18050f3f252c244e0f293a0e/msgspec-0.22.0-cp313-cp313-win_amd64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numcodecs_sdist_eeed77e4": {
+          "ncore_pip_deps_313_msgspec_sdist_0a13624a": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21517,17 +21889,17 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "numcodecs-0.15.1.tar.gz",
+              "filename": "msgspec-0.22.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numcodecs==0.15.1",
-              "sha256": "eeed77e4d6636641a2cc605fbc6078c7a8f2cc40f3dfa2b3f61e52e6091b04ff",
+              "requirement": "msgspec==0.22.0",
+              "sha256": "0a13624a4969159fe35d8c2a3d377b2b61bbd8585e327440d5e52725affcce38",
               "urls": [
-                "https://files.pythonhosted.org/packages/63/fc/bb532969eb8236984ba65e4f0079a7da885b8ac0ce1f0835decbb3938a62/numcodecs-0.15.1.tar.gz"
+                "https://files.pythonhosted.org/packages/d0/e6/6dcf9306ff3c5e486578f3bf29ed11dfbdbbc2a8bf0caf7e07d392887fda/msgspec-0.22.0.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_10_13_x86_64_2377da2d": {
+          "ncore_pip_deps_313_mypy_extensions_py3_none_any_1be4cccd": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21541,17 +21913,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-macosx_10_13_x86_64.whl",
+              "filename": "mypy_extensions-1.1.0-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "2377da2dd3ba2c1200956acbab2a358c83b8e1f8531191672d1cd6ad83250d53",
+              "requirement": "mypy-extensions==1.1.0",
+              "sha256": "1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505",
               "urls": [
-                "https://files.pythonhosted.org/packages/67/14/1c3ee0118a8fce08565a5d8482631608426a33af10a01077fada5dc7c119/numpy-2.5.4-cp313-cp313-macosx_10_13_x86_64.whl"
+                "https://files.pythonhosted.org/packages/79/7b/2c79738432f5c924bef5071f933bcc9efd0473bac3b4aa584a6f7c1c8df8/mypy_extensions-1.1.0-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_11_0_arm64_7415db95": {
+          "ncore_pip_deps_313_mypy_extensions_sdist_52e68efc": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21565,17 +21937,23 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-macosx_11_0_arm64.whl",
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "mypy_extensions-1.1.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "7415db95818b39ec475a5eea54d9e3b6bc83e3912158e46da3438cdce399804d",
+              "requirement": "mypy-extensions==1.1.0",
+              "sha256": "52e68efc3284861e772bbcd66823fde5ae21fd2fdb51c62a211403730b916558",
               "urls": [
-                "https://files.pythonhosted.org/packages/83/8c/b0ea9477fb1f0d4484bbc5cba21678cc9969704d8d7f3f158d1db35f8e14/numpy-2.5.4-cp313-cp313-macosx_11_0_arm64.whl"
+                "https://files.pythonhosted.org/packages/a2/6e/371856a3fb9d31ca8dac321cda606860fa4548858c0cc45d9d1d4ca2628b/mypy_extensions-1.1.0.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_arm64_6d6a71b9": {
+          "ncore_pip_deps_313_networkx_py3_none_any_e3fd2c13": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21589,17 +21967,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-macosx_14_0_arm64.whl",
+              "filename": "networkx-3.7-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "6d6a71b9d9a97c03633aa12565ef2825ffa036cc1d99cfd50dacf0f128af4fe2",
+              "requirement": "networkx==3.7",
+              "sha256": "e3fd2c13a7814cee3746340d8d7f8598a67f16a58bf47fb7f8793fab6efca1b0",
               "urls": [
-                "https://files.pythonhosted.org/packages/e2/84/6a3d75b3ba3dfe84ac0053450753d1e6d250a8bf80f66474cc46d1fb643f/numpy-2.5.4-cp313-cp313-macosx_14_0_arm64.whl"
+                "https://files.pythonhosted.org/packages/7e/cd/fe58041e9011f307c490e3e17dd48cc516448f7c698a3f2d9d9d65d7e6a8/networkx-3.7-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_x86_64_d8200f16": {
+          "ncore_pip_deps_313_networkx_sdist_fd77a511": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21613,17 +21991,23 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-macosx_14_0_x86_64.whl",
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "networkx-3.7.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "d8200f16437b289a5bb927c6e184eccc3e8389bc0070fea4cd5b9e13c1757959",
+              "requirement": "networkx==3.7",
+              "sha256": "fd77a511bd90f39f3d016351345b52cf5319b813bdca01de3f755d3cca62e96a",
               "urls": [
-                "https://files.pythonhosted.org/packages/61/18/bb993f267ca20b376e07092a16793a5b31ed3138751e9ba480011a14d742/numpy-2.5.4-cp313-cp313-macosx_14_0_x86_64.whl"
+                "https://files.pythonhosted.org/packages/dc/76/3af777226b63a5e64a6b36b1ec5855c14e2b94a37096d4760e595fc43511/networkx-3.7.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_aarch64_1c2e71b0": {
+          "ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_10_13_x86_64_de2c66db": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21637,17 +22021,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl",
+              "filename": "numcodecs-0.17.0-cp313-cp313-macosx_10_13_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "1c2e71b04c6cad90026e544501bbe0ab9290fa8a4d845e7e8c0d124fb429c988",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "de2c66db238e74e66fe9be7e02b7e0129b75d3f812d38e4019eb0102cc2dcdf0",
               "urls": [
-                "https://files.pythonhosted.org/packages/db/b6/135bb0953b61dc21c6cafa14b424ae666944e4899cf140e00c2b322a1a45/numpy-2.5.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
+                "https://files.pythonhosted.org/packages/11/f1/1d3d2bcb1240e5000f6647b5b0fd465b2b51ecef180bfa797a85df48cf2f/numcodecs-0.17.0-cp313-cp313-macosx_10_13_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_x86_64_6ffa0766": {
+          "ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_11_0_arm64_69b9b468": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21661,17 +22045,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
+              "filename": "numcodecs-0.17.0-cp313-cp313-macosx_11_0_arm64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "6ffa07666f8da0eef81d149934a626d0d95fbd6838432a33e66245423a9062c0",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "69b9b4685097c4d478a0c829debf4470555ec63e92cdd2c6b5f195460f1dc888",
               "urls": [
-                "https://files.pythonhosted.org/packages/da/24/3bd070f3269dc609d8f26b2643f62ef91bb415841c0b294805aaf7fe06da/numpy-2.5.4-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl"
+                "https://files.pythonhosted.org/packages/64/81/64e2472a8b3a9fa26bccfc7d5fa876770a9027bd5cd77e5b4a7b807a0785/numcodecs-0.17.0-cp313-cp313-macosx_11_0_arm64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_aarch64_2fa3328f": {
+          "ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_aarch64_7065b334": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21685,17 +22069,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-musllinux_1_2_aarch64.whl",
+              "filename": "numcodecs-0.17.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "2fa3328f784fc8277fc48026f6cad516f5c561c5d8e2e39b3c9e0c8f23223b34",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "7065b3349b73d54785aa89e00d0b97d80f664e9056757929d28151f9208dc04c",
               "urls": [
-                "https://files.pythonhosted.org/packages/c7/8e/9d15bd356b0a019c965312b1a3c6a727cac4cae5bc40045fbc12ce4cff9c/numpy-2.5.4-cp313-cp313-musllinux_1_2_aarch64.whl"
+                "https://files.pythonhosted.org/packages/25/ea/2ab25f7e674cf1e78f123c5c2689d8a7dc85475554af0619bcce05cb32a9/numcodecs-0.17.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_x86_64_b86966fb": {
+          "ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_x86_64_6c3342d9": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21709,17 +22093,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-musllinux_1_2_x86_64.whl",
+              "filename": "numcodecs-0.17.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "b86966fbe4ad7de710422175572bcdc75fdedadfb54bc6fab7deabccddd7780b",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "6c3342d91ed7cf59c1be84396edd364e936bb0ec9e366d24bb69689748d19625",
               "urls": [
-                "https://files.pythonhosted.org/packages/dc/fe/9d5b560db964f15871885f2250795d15945f8699e17ef90c0c2ff4c875b2/numpy-2.5.4-cp313-cp313-musllinux_1_2_x86_64.whl"
+                "https://files.pythonhosted.org/packages/9d/96/b3bf9a31978d936654a73f2bb1036b92b6515164f170092d162419eb771c/numcodecs-0.17.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_cp313_cp313_win_amd64_8b4d2fd2": {
+          "ncore_pip_deps_313_numcodecs_cp313_cp313_win_amd64_a854e9c8": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21733,17 +22117,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "numpy-2.5.4-cp313-cp313-win_amd64.whl",
+              "filename": "numcodecs-0.17.0-cp313-cp313-win_amd64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "8b4d2fd2d34e5f8c9235ee787de5631a37a28402b15cb80814df973d2be54129",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "a854e9c89f58eeeb2453f3c1637d1916797edb6eaff26bc186a6cdb09d187092",
               "urls": [
-                "https://files.pythonhosted.org/packages/90/8c/140a40398a66b4471211be1affdb6ed24c486d581bd28d07b7f2fcb69540/numpy-2.5.4-cp313-cp313-win_amd64.whl"
+                "https://files.pythonhosted.org/packages/5c/ec/47515bea31725376aa6f061c335326cc7437f863ab704b8e167e80734bfd/numcodecs-0.17.0-cp313-cp313-win_amd64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_numpy_sdist_9a94cf75": {
+          "ncore_pip_deps_313_numcodecs_sdist_e8db2e33": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21763,17 +22147,17 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "numpy-2.5.4.tar.gz",
+              "filename": "numcodecs-0.17.0.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "numpy==2.5.4",
-              "sha256": "9a94cf751c9ad8ebaa835bcd3d40dacf8534ad086b88c38029b65123c7999d2a",
+              "requirement": "numcodecs==0.17.0",
+              "sha256": "e8db2e337bdafd3bb5f891a2543b53b2b36a509ce9d587af2846db3715b6c8b9",
               "urls": [
-                "https://files.pythonhosted.org/packages/95/b0/c7453d0b6e2073c3264468b106ee1563750cecc910965e67357e3698c83e/numpy-2.5.4.tar.gz"
+                "https://files.pythonhosted.org/packages/dd/ec/260cdb6304868de6db14eb31064bd2735c0200bcb3331d6b4c9e9be02a03/numcodecs-0.17.0.tar.gz"
               ]
             }
           },
-          "ncore_pip_deps_313_opencv_python_headless_cp37_abi3_macosx_13_0_arm64_030ca5e0": {
+          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_10_13_x86_64_2377da2d": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21787,17 +22171,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "opencv_python_headless-5.0.0.93-cp37-abi3-macosx_13_0_arm64.whl",
+              "filename": "numpy-2.5.4-cp313-cp313-macosx_10_13_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "opencv-python-headless==5.0.0.93",
-              "sha256": "030ca5e0837a2963ab36ef896baa9767eb8d2b83353fb28af5a521e40dd8756f",
+              "requirement": "numpy==2.5.4",
+              "sha256": "2377da2dd3ba2c1200956acbab2a358c83b8e1f8531191672d1cd6ad83250d53",
               "urls": [
-                "https://files.pythonhosted.org/packages/53/7c/8c8097891c509d98cd128493835c95631c80be6a8f37ed9d25716c2e16f1/opencv_python_headless-5.0.0.93-cp37-abi3-macosx_13_0_arm64.whl"
+                "https://files.pythonhosted.org/packages/67/14/1c3ee0118a8fce08565a5d8482631608426a33af10a01077fada5dc7c119/numpy-2.5.4-cp313-cp313-macosx_10_13_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_opencv_python_headless_cp37_abi3_macosx_14_0_x86_64_1e55af3a": {
+          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_11_0_arm64_7415db95": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -21811,7 +22195,253 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "opencv_python_headless-5.0.0.93-cp37-abi3-macosx_14_0_x86_64.whl",
+              "filename": "numpy-2.5.4-cp313-cp313-macosx_11_0_arm64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "7415db95818b39ec475a5eea54d9e3b6bc83e3912158e46da3438cdce399804d",
+              "urls": [
+                "https://files.pythonhosted.org/packages/83/8c/b0ea9477fb1f0d4484bbc5cba21678cc9969704d8d7f3f158d1db35f8e14/numpy-2.5.4-cp313-cp313-macosx_11_0_arm64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_arm64_6d6a71b9": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-macosx_14_0_arm64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "6d6a71b9d9a97c03633aa12565ef2825ffa036cc1d99cfd50dacf0f128af4fe2",
+              "urls": [
+                "https://files.pythonhosted.org/packages/e2/84/6a3d75b3ba3dfe84ac0053450753d1e6d250a8bf80f66474cc46d1fb643f/numpy-2.5.4-cp313-cp313-macosx_14_0_arm64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_x86_64_d8200f16": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-macosx_14_0_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "d8200f16437b289a5bb927c6e184eccc3e8389bc0070fea4cd5b9e13c1757959",
+              "urls": [
+                "https://files.pythonhosted.org/packages/61/18/bb993f267ca20b376e07092a16793a5b31ed3138751e9ba480011a14d742/numpy-2.5.4-cp313-cp313-macosx_14_0_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_aarch64_1c2e71b0": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "1c2e71b04c6cad90026e544501bbe0ab9290fa8a4d845e7e8c0d124fb429c988",
+              "urls": [
+                "https://files.pythonhosted.org/packages/db/b6/135bb0953b61dc21c6cafa14b424ae666944e4899cf140e00c2b322a1a45/numpy-2.5.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_x86_64_6ffa0766": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "6ffa07666f8da0eef81d149934a626d0d95fbd6838432a33e66245423a9062c0",
+              "urls": [
+                "https://files.pythonhosted.org/packages/da/24/3bd070f3269dc609d8f26b2643f62ef91bb415841c0b294805aaf7fe06da/numpy-2.5.4-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_aarch64_2fa3328f": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-musllinux_1_2_aarch64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "2fa3328f784fc8277fc48026f6cad516f5c561c5d8e2e39b3c9e0c8f23223b34",
+              "urls": [
+                "https://files.pythonhosted.org/packages/c7/8e/9d15bd356b0a019c965312b1a3c6a727cac4cae5bc40045fbc12ce4cff9c/numpy-2.5.4-cp313-cp313-musllinux_1_2_aarch64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_x86_64_b86966fb": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-musllinux_1_2_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "b86966fbe4ad7de710422175572bcdc75fdedadfb54bc6fab7deabccddd7780b",
+              "urls": [
+                "https://files.pythonhosted.org/packages/dc/fe/9d5b560db964f15871885f2250795d15945f8699e17ef90c0c2ff4c875b2/numpy-2.5.4-cp313-cp313-musllinux_1_2_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_cp313_cp313_win_amd64_8b4d2fd2": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "numpy-2.5.4-cp313-cp313-win_amd64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "8b4d2fd2d34e5f8c9235ee787de5631a37a28402b15cb80814df973d2be54129",
+              "urls": [
+                "https://files.pythonhosted.org/packages/90/8c/140a40398a66b4471211be1affdb6ed24c486d581bd28d07b7f2fcb69540/numpy-2.5.4-cp313-cp313-win_amd64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_numpy_sdist_9a94cf75": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "numpy-2.5.4.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "numpy==2.5.4",
+              "sha256": "9a94cf751c9ad8ebaa835bcd3d40dacf8534ad086b88c38029b65123c7999d2a",
+              "urls": [
+                "https://files.pythonhosted.org/packages/95/b0/c7453d0b6e2073c3264468b106ee1563750cecc910965e67357e3698c83e/numpy-2.5.4.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_opencv_python_headless_cp37_abi3_macosx_13_0_arm64_030ca5e0": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "opencv_python_headless-5.0.0.93-cp37-abi3-macosx_13_0_arm64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "opencv-python-headless==5.0.0.93",
+              "sha256": "030ca5e0837a2963ab36ef896baa9767eb8d2b83353fb28af5a521e40dd8756f",
+              "urls": [
+                "https://files.pythonhosted.org/packages/53/7c/8c8097891c509d98cd128493835c95631c80be6a8f37ed9d25716c2e16f1/opencv_python_headless-5.0.0.93-cp37-abi3-macosx_13_0_arm64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_opencv_python_headless_cp37_abi3_macosx_14_0_x86_64_1e55af3a": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "opencv_python_headless-5.0.0.93-cp37-abi3-macosx_14_0_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
               "requirement": "opencv-python-headless==5.0.0.93",
@@ -22294,14 +22924,254 @@
               "filename": "pillow-12.3.0-cp313-cp313-win_amd64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pillow==12.3.0",
-              "sha256": "1cca606cd25738df4ed873d5ad46bbdb3d83b5cbca291f6b4ff13a4df6b0bbe8",
+              "requirement": "pillow==12.3.0",
+              "sha256": "1cca606cd25738df4ed873d5ad46bbdb3d83b5cbca291f6b4ff13a4df6b0bbe8",
+              "urls": [
+                "https://files.pythonhosted.org/packages/a6/9b/7a58e61d62be561da3a356fe2384d4059a6345fc130e23ef1c36a5b81d24/pillow-12.3.0-cp313-cp313-win_amd64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pillow_sdist_3b8182a7": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "pillow-12.3.0.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pillow==12.3.0",
+              "sha256": "3b8182a766685eaa002637e28b4ec8d6b18819a0c71f579bf0dbaa5830297cce",
+              "urls": [
+                "https://files.pythonhosted.org/packages/1c/3d/bb7fca845737cf9d7dbde16ed1843984665ff2e0a518f5db43e77ec540b9/pillow-12.3.0.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pluggy_py3_none_any_e920276d": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "pluggy-1.6.0-py3-none-any.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pluggy==1.6.0",
+              "sha256": "e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746",
+              "urls": [
+                "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pluggy_sdist_7dcc130b": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "pluggy-1.6.0.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pluggy==1.6.0",
+              "sha256": "7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3",
+              "urls": [
+                "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pygments_py3_none_any_2363c69b": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "pygments-2.21.0-py3-none-any.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pygments==2.21.0",
+              "sha256": "2363c69b61c4a97c838da3b130dcd6468f4848992b21a82f2a63ec34377137d9",
+              "urls": [
+                "https://files.pythonhosted.org/packages/71/46/17f022dd3e953bf20a04a028a21ec746d942f8d2af30fa0f124fa0e6a684/pygments-2.21.0-py3-none-any.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pygments_sdist_610ca751": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "pygments-2.21.0.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pygments==2.21.0",
+              "sha256": "610ca751c9bc2492b38eb9a38a7fbc93edbbb2d7182edaf34e66ae493dee5c8c",
+              "urls": [
+                "https://files.pythonhosted.org/packages/49/2e/ced460408999b33da6b31b0021b0f37d329e202d4169aeb164493778f25b/pygments-2.21.0.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pytest_py3_none_any_37a86b45": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "pytest-9.1.1-py3-none-any.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pytest==9.1.1",
+              "sha256": "37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c",
+              "urls": [
+                "https://files.pythonhosted.org/packages/24/25/1de2678b631f5a49215c6c96fff41ba892b0a34df68d6d80292b1b48aa7f/pytest-9.1.1-py3-none-any.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pytest_sdist_1088fbde": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "extra_pip_args": [
+                "--index-url",
+                "https://pypi.org/simple",
+                "--find-links",
+                "https://download.pytorch.org/whl/cpu/torch"
+              ],
+              "filename": "pytest-9.1.1.tar.gz",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pytest==9.1.1",
+              "sha256": "1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313",
+              "urls": [
+                "https://files.pythonhosted.org/packages/e4/47/b9efed96c114afcfa3c9d3fe98a76a1d14c74a9e266d397cf6eb64be5e01/pytest-9.1.1.tar.gz"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_macosx_10_13_x86_64_8da9669d": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp313-cp313-macosx_10_13_x86_64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8",
+              "urls": [
+                "https://files.pythonhosted.org/packages/d1/11/0fd08f8192109f7169db964b5707a2f1e8b745d4e239b784a5a1dd80d1db/pyyaml-6.0.3-cp313-cp313-macosx_10_13_x86_64.whl"
+              ]
+            }
+          },
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_macosx_11_0_arm64_2283a07e": {
+            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
+            "attributes": {
+              "dep_template": "@ncore_pip_deps//{name}:{target}",
+              "experimental_target_platforms": [
+                "cp313_linux_aarch64",
+                "cp313_linux_arm",
+                "cp313_linux_ppc",
+                "cp313_linux_s390x",
+                "cp313_linux_x86_64",
+                "cp313_osx_aarch64",
+                "cp313_osx_x86_64",
+                "cp313_windows_x86_64"
+              ],
+              "filename": "pyyaml-6.0.3-cp313-cp313-macosx_11_0_arm64.whl",
+              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
+              "repo": "ncore_pip_deps_313",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1",
               "urls": [
-                "https://files.pythonhosted.org/packages/a6/9b/7a58e61d62be561da3a356fe2384d4059a6345fc130e23ef1c36a5b81d24/pillow-12.3.0-cp313-cp313-win_amd64.whl"
+                "https://files.pythonhosted.org/packages/b1/16/95309993f1d3748cd644e02e38b75d50cbc0d9561d21f390a76242ce073f/pyyaml-6.0.3-cp313-cp313-macosx_11_0_arm64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pillow_sdist_3b8182a7": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_aarch64_ee292290": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22315,23 +23185,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "pillow-12.3.0.tar.gz",
+              "filename": "pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pillow==12.3.0",
-              "sha256": "3b8182a766685eaa002637e28b4ec8d6b18819a0c71f579bf0dbaa5830297cce",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c",
               "urls": [
-                "https://files.pythonhosted.org/packages/1c/3d/bb7fca845737cf9d7dbde16ed1843984665ff2e0a518f5db43e77ec540b9/pillow-12.3.0.tar.gz"
+                "https://files.pythonhosted.org/packages/50/31/b20f376d3f810b9b2371e72ef5adb33879b25edb7a6d072cb7ca0c486398/pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pluggy_py3_none_any_e920276d": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_s390x_a33284e2": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22345,17 +23209,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "pluggy-1.6.0-py3-none-any.whl",
+              "filename": "pyyaml-6.0.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pluggy==1.6.0",
-              "sha256": "e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5",
               "urls": [
-                "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/49/1e/a55ca81e949270d5d4432fbbd19dfea5321eda7c41a849d443dc92fd1ff7/pyyaml-6.0.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pluggy_sdist_7dcc130b": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_x86_64_0f29edc4": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22369,23 +23233,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "pluggy-1.6.0.tar.gz",
+              "filename": "pyyaml-6.0.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pluggy==1.6.0",
-              "sha256": "7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6",
               "urls": [
-                "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz"
+                "https://files.pythonhosted.org/packages/74/27/e5b8f34d02d9995b80abcef563ea1f8b56d20134d8f4e5e81733b1feceb2/pyyaml-6.0.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pygments_py3_none_any_2363c69b": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_musllinux_1_2_aarch64_f7057c9a": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22399,17 +23257,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "pygments-2.21.0-py3-none-any.whl",
+              "filename": "pyyaml-6.0.3-cp313-cp313-musllinux_1_2_aarch64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pygments==2.21.0",
-              "sha256": "2363c69b61c4a97c838da3b130dcd6468f4848992b21a82f2a63ec34377137d9",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6",
               "urls": [
-                "https://files.pythonhosted.org/packages/71/46/17f022dd3e953bf20a04a028a21ec746d942f8d2af30fa0f124fa0e6a684/pygments-2.21.0-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/f9/11/ba845c23988798f40e52ba45f34849aa8a1f2d4af4b798588010792ebad6/pyyaml-6.0.3-cp313-cp313-musllinux_1_2_aarch64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pygments_sdist_610ca751": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_musllinux_1_2_x86_64_eda16858": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22423,23 +23281,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "pygments-2.21.0.tar.gz",
+              "filename": "pyyaml-6.0.3-cp313-cp313-musllinux_1_2_x86_64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pygments==2.21.0",
-              "sha256": "610ca751c9bc2492b38eb9a38a7fbc93edbbb2d7182edaf34e66ae493dee5c8c",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be",
               "urls": [
-                "https://files.pythonhosted.org/packages/49/2e/ced460408999b33da6b31b0021b0f37d329e202d4169aeb164493778f25b/pygments-2.21.0.tar.gz"
+                "https://files.pythonhosted.org/packages/3d/e0/7966e1a7bfc0a45bf0a7fb6b98ea03fc9b8d84fa7f2229e9659680b69ee3/pyyaml-6.0.3-cp313-cp313-musllinux_1_2_x86_64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pytest_py3_none_any_37a86b45": {
+          "ncore_pip_deps_313_pyyaml_cp313_cp313_win_amd64_79005a0d": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22453,17 +23305,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "pytest-9.1.1-py3-none-any.whl",
+              "filename": "pyyaml-6.0.3-cp313-cp313-win_amd64.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pytest==9.1.1",
-              "sha256": "37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c",
               "urls": [
-                "https://files.pythonhosted.org/packages/24/25/1de2678b631f5a49215c6c96fff41ba892b0a34df68d6d80292b1b48aa7f/pytest-9.1.1-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/97/c9/39d5b874e8b28845e4ec2202b5da735d0199dbe5b8fb85f91398814a9a46/pyyaml-6.0.3-cp313-cp313-win_amd64.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_pytest_sdist_1088fbde": {
+          "ncore_pip_deps_313_pyyaml_sdist_d7662337": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -22483,13 +23335,13 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "pytest-9.1.1.tar.gz",
+              "filename": "pyyaml-6.0.3.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "pytest==9.1.1",
-              "sha256": "1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313",
+              "requirement": "pyyaml==6.0.3",
+              "sha256": "d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f",
               "urls": [
-                "https://files.pythonhosted.org/packages/e4/47/b9efed96c114afcfa3c9d3fe98a76a1d14c74a9e266d397cf6eb64be5e01/pytest-9.1.1.tar.gz"
+                "https://files.pythonhosted.org/packages/05/8e/961c0007c59b8dd7729d542c61a4d537767a59645b82a0b521206e1e25c2/pyyaml-6.0.3.tar.gz"
               ]
             }
           },
@@ -23027,229 +23879,7 @@
               ]
             }
           },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_macosx_10_13_x86_64_b312b3cc": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-macosx_10_13_x86_64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "b312b3cc87951faaed3cfef984d768ee8bee7f935d9cc929aaa9946b0b96a98c",
-              "urls": [
-                "https://files.pythonhosted.org/packages/d6/4b/cc7bb5668f7ddc0e73e236e96a0c06cab8fddfca9c53538c9dffac62db6f/wrapt-2.5.0-cp313-cp313-macosx_10_13_x86_64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_macosx_11_0_arm64_c57ddae2": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-macosx_11_0_arm64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "c57ddae24cf72eb6bd18112638a987cafe6109d90f2df111e6934362cc03ac1a",
-              "urls": [
-                "https://files.pythonhosted.org/packages/4a/13/5d15ef0e2f42d5f084930dc4863e6e52c160c28301c8780aae170c58421c/wrapt-2.5.0-cp313-cp313-macosx_11_0_arm64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_manylinux_2_17_aarch64_6058e12e": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "6058e12e9caa33468f9a36fb88c15a4bb30a479f997b37834b83abdbf062f264",
-              "urls": [
-                "https://files.pythonhosted.org/packages/a4/f9/47ae1d7ef325c3f6c81ae3c1fb4a3fef9d98c8025ed676c0bfc1550903ce/wrapt-2.5.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_manylinux_2_5_x86_64_b95a6eca": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "b95a6eca3b927853529eea958310563c83140ae8451dd5dc4399c7da385dc4f3",
-              "urls": [
-                "https://files.pythonhosted.org/packages/1e/02/c7174e78b0c38bb279b2d3c25a6bd7fb9d3b0200c3e5a8fad084e7cc3e85/wrapt-2.5.0-cp313-cp313-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_musllinux_1_2_aarch64_f4ef4935": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-musllinux_1_2_aarch64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "f4ef4935962f7029b2058a99f1a47ccbffc3be919dddb3becb6c2c48eac3d9f0",
-              "urls": [
-                "https://files.pythonhosted.org/packages/41/45/fc252bda5aa1ca01bc838d3b108778e786a2a13d0c52fd17c5f6179aa246/wrapt-2.5.0-cp313-cp313-musllinux_1_2_aarch64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_musllinux_1_2_x86_64_a346408f": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-musllinux_1_2_x86_64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "a346408f19b6d589bf029f25f65c0b4cdeed6302ef8f40da4e5d1552d22dc037",
-              "urls": [
-                "https://files.pythonhosted.org/packages/81/8e/2b823fded8c3b815408c58633929812eacd29d824fe57548b7868c4ee422/wrapt-2.5.0-cp313-cp313-musllinux_1_2_x86_64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_cp313_cp313_win_amd64_77f0a74f": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-cp313-cp313-win_amd64.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "77f0a74ff6f6cf89f5b673a732d5afe1911a6e6b1c017260836fdfdf85518dc1",
-              "urls": [
-                "https://files.pythonhosted.org/packages/ce/9a/51d95640e01d0ebdd04a7223755f076e4936b0c124ce99bb01a12b53e66c/wrapt-2.5.0-cp313-cp313-win_amd64.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_py3_none_any_107eea1a": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "filename": "wrapt-2.5.0-py3-none-any.whl",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "107eea1a511e98a3a5033b0c2cb403fbb37f05dee6ac1fb85c0460d311ec278c",
-              "urls": [
-                "https://files.pythonhosted.org/packages/87/7d/5ed859fad4b5eddd598a846150aaab2703730ed4886c5c5e03b0df0cfdd5/wrapt-2.5.0-py3-none-any.whl"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_wrapt_sdist_c48cdb6c": {
-            "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
-            "attributes": {
-              "dep_template": "@ncore_pip_deps//{name}:{target}",
-              "experimental_target_platforms": [
-                "cp313_linux_aarch64",
-                "cp313_linux_arm",
-                "cp313_linux_ppc",
-                "cp313_linux_s390x",
-                "cp313_linux_x86_64",
-                "cp313_osx_aarch64",
-                "cp313_osx_x86_64",
-                "cp313_windows_x86_64"
-              ],
-              "extra_pip_args": [
-                "--index-url",
-                "https://pypi.org/simple",
-                "--find-links",
-                "https://download.pytorch.org/whl/cpu/torch"
-              ],
-              "filename": "wrapt-2.5.0.tar.gz",
-              "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
-              "repo": "ncore_pip_deps_313",
-              "requirement": "wrapt==2.5.0",
-              "sha256": "c48cdb6c904dca76d9915a579e4a5fab6b0c25f650c1019ce78a78effaf7a345",
-              "urls": [
-                "https://files.pythonhosted.org/packages/3e/d2/a254a26d8ceaea87e0eee2e89fcfe53ddc1858418647493bb2937549ab6f/wrapt-2.5.0.tar.gz"
-              ]
-            }
-          },
-          "ncore_pip_deps_313_zarr_py3_none_any_ac3dc403": {
+          "ncore_pip_deps_313_zarr_py3_none_any_38b54057": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -23263,17 +23893,17 @@
                 "cp313_osx_x86_64",
                 "cp313_windows_x86_64"
               ],
-              "filename": "zarr-2.18.7-py3-none-any.whl",
+              "filename": "zarr-3.4.1-py3-none-any.whl",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "zarr==2.18.7",
-              "sha256": "ac3dc4033e9ae4e9d7b5e27c97ea3eaf1003cc0a07f010bd83d5134bf8c4b223",
+              "requirement": "zarr==3.4.1",
+              "sha256": "38b540578a119352bdce02a720d7bfd99846e77f0728c58b1bab3d1f547526a0",
               "urls": [
-                "https://files.pythonhosted.org/packages/5e/d8/9ffd8c237b3559945bb52103cf0eed64ea098f7b7f573f8d2962ef27b4b2/zarr-2.18.7-py3-none-any.whl"
+                "https://files.pythonhosted.org/packages/a6/82/0dbc9bc77b49cfb9268dc2b2b1dcd04346c8c7ed52e272b76a628a941656/zarr-3.4.1-py3-none-any.whl"
               ]
             }
           },
-          "ncore_pip_deps_313_zarr_sdist_b2b8f66f": {
+          "ncore_pip_deps_313_zarr_sdist_b34bda11": {
             "repoRuleId": "@@rules_python+//python/private/pypi:whl_library.bzl%whl_library",
             "attributes": {
               "dep_template": "@ncore_pip_deps//{name}:{target}",
@@ -23293,13 +23923,13 @@
                 "--find-links",
                 "https://download.pytorch.org/whl/cpu/torch"
               ],
-              "filename": "zarr-2.18.7.tar.gz",
+              "filename": "zarr-3.4.1.tar.gz",
               "python_interpreter_target": "@@rules_python++python+python_3_13_host//:python",
               "repo": "ncore_pip_deps_313",
-              "requirement": "zarr==2.18.7",
-              "sha256": "b2b8f66f14dac4af66b180d2338819981b981f70e196c9a66e6bfaa9e59572f5",
+              "requirement": "zarr==3.4.1",
+              "sha256": "b34bda11ceb199c81ee78ecd42cd02f46c7f67a6d8a1e9bc501cafd5a1795356",
               "urls": [
-                "https://files.pythonhosted.org/packages/da/1d/01cf9e3ab2d85190278efc3fca9f68563de35ae30ee59e7640e3af98abe3/zarr-2.18.7.tar.gz"
+                "https://files.pythonhosted.org/packages/3e/62/e8a36a4b65f01499c7aefcf103aabba0c37c018bf135fbf179f6ed2f01a0/zarr-3.4.1.tar.gz"
               ]
             }
           },
@@ -27715,7 +28345,7 @@
                 "accessible_pygments": "{\"ncore_pip_deps_311_accessible_pygments_py3_none_any_88ae3211\":[{\"filename\":\"accessible_pygments-0.0.5-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_accessible_pygments_sdist_40918d3e\":[{\"filename\":\"accessible_pygments-0.0.5.tar.gz\",\"version\":\"3.11\"}]}",
                 "alabaster": "{\"ncore_pip_deps_311_alabaster_py3_none_any_b46733c0\":[{\"filename\":\"alabaster-0.7.16-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_alabaster_sdist_75a8b99c\":[{\"filename\":\"alabaster-0.7.16.tar.gz\",\"version\":\"3.11\"}]}",
                 "anyio": "{\"ncore_pip_deps_311_anyio_py3_none_any_d4058288\":[{\"filename\":\"anyio-4.12.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_anyio_sdist_41cfcc3a\":[{\"filename\":\"anyio-4.12.1.tar.gz\",\"version\":\"3.11\"}]}",
-                "asciitree": "{\"ncore_pip_deps_311_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.8\"}]}",
+                "asciitree": "{\"ncore_pip_deps_312_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_38_asciitree_sdist_4aa4b9b6\":[{\"filename\":\"asciitree-0.3.3.tar.gz\",\"version\":\"3.8\"}]}",
                 "asttokens": "{\"ncore_pip_deps_311_asttokens_py3_none_any_15a3ebc0\":[{\"filename\":\"asttokens-3.0.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_asttokens_sdist_71a4ee5d\":[{\"filename\":\"asttokens-3.0.1.tar.gz\",\"version\":\"3.11\"}]}",
                 "astunparse": "{\"ncore_pip_deps_311_astunparse_py2_none_any_c2652417\":[{\"filename\":\"astunparse-1.6.3-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_astunparse_sdist_5ad93a84\":[{\"filename\":\"astunparse-1.6.3.tar.gz\",\"version\":\"3.11\"}]}",
                 "attrs": "{\"ncore_pip_deps_311_attrs_py3_none_any_c75a69e2\":[{\"filename\":\"attrs-25.1.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_attrs_sdist_1c97078a\":[{\"filename\":\"attrs-25.1.0.tar.gz\",\"version\":\"3.11\"}]}",
@@ -27734,15 +28364,16 @@
                 "debugpy": "{\"ncore_pip_deps_311_debugpy_cp311_cp311_macosx_14_0_universal2_36f48298\":[{\"filename\":\"debugpy-1.8.12-cp311-cp311-macosx_14_0_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_debugpy_cp311_cp311_manylinux_2_5_x86_64_a28ed481\":[{\"filename\":\"debugpy-1.8.12-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_debugpy_cp311_cp311_win_amd64_4703575b\":[{\"filename\":\"debugpy-1.8.12-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_debugpy_py2_none_any_274b6a20\":[{\"filename\":\"debugpy-1.8.12-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_debugpy_sdist_646530b0\":[{\"filename\":\"debugpy-1.8.12.tar.gz\",\"version\":\"3.11\"}]}",
                 "decorator": "{\"ncore_pip_deps_311_decorator_py3_none_any_d316bb41\":[{\"filename\":\"decorator-5.2.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_decorator_sdist_65f26614\":[{\"filename\":\"decorator-5.2.1.tar.gz\",\"version\":\"3.11\"}]}",
                 "defusedxml": "{\"ncore_pip_deps_311_defusedxml_py2_none_any_a352e7e4\":[{\"filename\":\"defusedxml-0.7.1-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_defusedxml_sdist_1bb3032d\":[{\"filename\":\"defusedxml-0.7.1.tar.gz\",\"version\":\"3.11\"}]}",
-                "deprecated": "{\"ncore_pip_deps_311_deprecated_py2_none_any_bd501178\":[{\"filename\":\"Deprecated-1.2.18-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_deprecated_sdist_422b6f6d\":[{\"filename\":\"deprecated-1.2.18.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_deprecated_py3_none_any_58204cf4\":[{\"filename\":\"deprecated-3.0.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_deprecated_sdist_16850204\":[{\"filename\":\"deprecated-3.0.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_deprecated_py3_none_any_58204cf4\":[{\"filename\":\"deprecated-3.0.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_deprecated_sdist_16850204\":[{\"filename\":\"deprecated-3.0.0.tar.gz\",\"version\":\"3.13\"}]}",
+                "deprecated": "{\"ncore_pip_deps_311_deprecated_py2_none_any_bd501178\":[{\"filename\":\"Deprecated-1.2.18-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_deprecated_sdist_422b6f6d\":[{\"filename\":\"deprecated-1.2.18.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_deprecated_py3_none_any_58204cf4\":[{\"filename\":\"deprecated-3.0.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_deprecated_sdist_16850204\":[{\"filename\":\"deprecated-3.0.0.tar.gz\",\"version\":\"3.12\"}]}",
                 "descartes": "{\"ncore_pip_deps_311_descartes_py2_none_any_b7e412e7\":[{\"filename\":\"descartes-1.1.0-py2-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_descartes_sdist_135a5021\":[{\"filename\":\"descartes-1.1.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "docutils": "{\"ncore_pip_deps_311_docutils_py3_none_any_d0013f54\":[{\"filename\":\"docutils-0.22.4-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_docutils_sdist_4db53b1f\":[{\"filename\":\"docutils-0.22.4.tar.gz\",\"version\":\"3.11\"}]}",
+                "donfig": "{\"ncore_pip_deps_311_donfig_py3_none_any_2a3175ce\":[{\"filename\":\"donfig-0.8.1.post1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_donfig_sdist_3bef3413\":[{\"filename\":\"donfig-0.8.1.post1.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_313_donfig_py3_none_any_2a3175ce\":[{\"filename\":\"donfig-0.8.1.post1-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_donfig_sdist_3bef3413\":[{\"filename\":\"donfig-0.8.1.post1.tar.gz\",\"version\":\"3.13\"}]}",
                 "dracopy": "{\"ncore_pip_deps_311_dracopy_cp311_cp311_macosx_10_9_x86_64_c4131151\":[{\"filename\":\"dracopy-2.0.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_dracopy_cp311_cp311_macosx_11_0_arm64_e16575ab\":[{\"filename\":\"dracopy-2.0.0-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_dracopy_cp311_cp311_manylinux_2_24_aarch64_418964da\":[{\"filename\":\"dracopy-2.0.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_dracopy_cp311_cp311_manylinux_2_24_x86_64_fa9a3215\":[{\"filename\":\"dracopy-2.0.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_dracopy_cp311_cp311_win_amd64_106c968f\":[{\"filename\":\"dracopy-2.0.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}]}",
                 "embreex": "{\"ncore_pip_deps_311_embreex_cp311_cp311_macosx_10_9_x86_64_8154ab34\":[{\"filename\":\"embreex-2.17.7.post7-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_embreex_cp311_cp311_manylinux_2_12_x86_64_c4775a4d\":[{\"filename\":\"embreex-2.17.7.post7-cp311-cp311-manylinux2010_x86_64.manylinux_2_12_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_embreex_cp311_cp311_win_amd64_a1157976\":[{\"filename\":\"embreex-2.17.7.post7-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}]}",
                 "entrypoints": "{\"ncore_pip_deps_38_entrypoints_py3_none_any_f174b5ff\":[{\"filename\":\"entrypoints-0.4-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_entrypoints_sdist_b706edda\":[{\"filename\":\"entrypoints-0.4.tar.gz\",\"version\":\"3.8\"}]}",
                 "exceptiongroup": "{\"ncore_pip_deps_38_exceptiongroup_py3_none_any_34328066\":[{\"filename\":\"exceptiongroup-1.1.3-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_exceptiongroup_sdist_097acd85\":[{\"filename\":\"exceptiongroup-1.1.3.tar.gz\",\"version\":\"3.8\"}]}",
                 "executing": "{\"ncore_pip_deps_311_executing_py2_none_any_11387150\":[{\"filename\":\"executing-2.2.0-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_executing_sdist_5d108c02\":[{\"filename\":\"executing-2.2.0.tar.gz\",\"version\":\"3.11\"}]}",
-                "fasteners": "{\"ncore_pip_deps_311_fasteners_py3_none_any_758819cb\":[{\"filename\":\"fasteners-0.19-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fasteners_sdist_b4f37c3a\":[{\"filename\":\"fasteners-0.19.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_fasteners_py3_none_any_9422c40d\":[{\"filename\":\"fasteners-0.20-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_fasteners_sdist_55dce879\":[{\"filename\":\"fasteners-0.20.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_fasteners_py3_none_any_9422c40d\":[{\"filename\":\"fasteners-0.20-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_fasteners_sdist_55dce879\":[{\"filename\":\"fasteners-0.20.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_fasteners_py3_none_any_1d4caf5f\":[{\"filename\":\"fasteners-0.18-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_fasteners_sdist_cb7c13ef\":[{\"filename\":\"fasteners-0.18.tar.gz\",\"version\":\"3.8\"}]}",
+                "fasteners": "{\"ncore_pip_deps_312_fasteners_py3_none_any_9422c40d\":[{\"filename\":\"fasteners-0.20-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_fasteners_sdist_55dce879\":[{\"filename\":\"fasteners-0.20.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_38_fasteners_py3_none_any_1d4caf5f\":[{\"filename\":\"fasteners-0.18-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_fasteners_sdist_cb7c13ef\":[{\"filename\":\"fasteners-0.18.tar.gz\",\"version\":\"3.8\"}]}",
                 "fastjsonschema": "{\"ncore_pip_deps_311_fastjsonschema_py3_none_any_c9e5b7e9\":[{\"filename\":\"fastjsonschema-2.21.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fastjsonschema_sdist_794d4f0a\":[{\"filename\":\"fastjsonschema-2.21.1.tar.gz\",\"version\":\"3.11\"}]}",
                 "filelock": "{\"ncore_pip_deps_311_filelock_py3_none_any_96f5f634\":[{\"filename\":\"filelock-3.29.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_filelock_sdist_69974355\":[{\"filename\":\"filelock-3.29.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_filelock_py3_none_any_2ce9818e\":[{\"filename\":\"filelock-4.1.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_filelock_sdist_ad7f724a\":[{\"filename\":\"filelock-4.1.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_filelock_py3_none_any_2ce9818e\":[{\"filename\":\"filelock-4.1.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_filelock_sdist_ad7f724a\":[{\"filename\":\"filelock-4.1.0.tar.gz\",\"version\":\"3.13\"}]}",
                 "fire": "{\"ncore_pip_deps_311_fire_py3_none_any_e43fd8a5\":[{\"filename\":\"fire-0.7.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fire_sdist_3b208f05\":[{\"filename\":\"fire-0.7.1.tar.gz\",\"version\":\"3.11\"}]}",
@@ -27750,6 +28381,7 @@
                 "fonttools": "{\"ncore_pip_deps_311_fonttools_cp311_cp311_macosx_10_9_universal2_7ef04bc7\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-macosx_10_9_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_macosx_10_9_x86_64_ffda9b8c\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_manylinux_2_17_aarch64_e2e993e8\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_manylinux_2_17_x86_64_003548ea\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_musllinux_1_2_aarch64_bd982582\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_musllinux_1_2_x86_64_b23d30a2\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_cp311_cp311_win_amd64_14a3e3e6\":[{\"filename\":\"fonttools-4.56.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_py3_none_any_1088182f\":[{\"filename\":\"fonttools-4.56.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fonttools_sdist_a114d156\":[{\"filename\":\"fonttools-4.56.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "fsspec": "{\"ncore_pip_deps_311_fsspec_py3_none_any_7c771235\":[{\"filename\":\"fsspec-2025.10.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_fsspec_sdist_b6789427\":[{\"filename\":\"fsspec-2025.10.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_fsspec_py3_none_any_8dd6e646\":[{\"filename\":\"fsspec-2026.9.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_fsspec_sdist_0f081479\":[{\"filename\":\"fsspec-2026.9.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_fsspec_py3_none_any_8dd6e646\":[{\"filename\":\"fsspec-2026.9.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_fsspec_sdist_0f081479\":[{\"filename\":\"fsspec-2026.9.0.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_fsspec_py3_none_any_efb87af3\":[{\"filename\":\"fsspec-2025.3.0-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_fsspec_sdist_a935fd1e\":[{\"filename\":\"fsspec-2025.3.0.tar.gz\",\"version\":\"3.8\"}]}",
                 "gast": "{\"ncore_pip_deps_311_gast_py3_none_any_52b18231\":[{\"filename\":\"gast-0.6.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_gast_sdist_88fc5300\":[{\"filename\":\"gast-0.6.0.tar.gz\",\"version\":\"3.11\"}]}",
+                "google_crc32c": "{\"ncore_pip_deps_311_google_crc32c_cp311_cp311_macosx_12_0_arm64_457d0d9a\":[{\"filename\":\"google_crc32c-1.9.0-cp311-cp311-macosx_12_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_crc32c_cp311_cp311_macosx_12_0_x86_64_ccfe4002\":[{\"filename\":\"google_crc32c-1.9.0-cp311-cp311-macosx_12_0_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_crc32c_cp311_cp311_manylinux_2_17_aarch64_fbef61a3\":[{\"filename\":\"google_crc32c-1.9.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_crc32c_cp311_cp311_manylinux_2_17_x86_64_86764b99\":[{\"filename\":\"google_crc32c-1.9.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_crc32c_cp311_cp311_win_amd64_43a2dc26\":[{\"filename\":\"google_crc32c-1.9.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_crc32c_sdist_7b8c84c3\":[{\"filename\":\"google_crc32c-1.9.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_313_google_crc32c_cp313_cp313_macosx_12_0_arm64_f1dc17d9\":[{\"filename\":\"google_crc32c-1.9.0-cp313-cp313-macosx_12_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_google_crc32c_cp313_cp313_macosx_12_0_x86_64_f894a287\":[{\"filename\":\"google_crc32c-1.9.0-cp313-cp313-macosx_12_0_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_google_crc32c_cp313_cp313_manylinux_2_17_aarch64_4488f155\":[{\"filename\":\"google_crc32c-1.9.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_google_crc32c_cp313_cp313_manylinux_2_17_x86_64_0568b17e\":[{\"filename\":\"google_crc32c-1.9.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_google_crc32c_cp313_cp313_win_amd64_8583ec21\":[{\"filename\":\"google_crc32c-1.9.0-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_google_crc32c_sdist_7b8c84c3\":[{\"filename\":\"google_crc32c-1.9.0.tar.gz\",\"version\":\"3.13\"}]}",
                 "google_pasta": "{\"ncore_pip_deps_311_google_pasta_py3_none_any_b3248279\":[{\"filename\":\"google_pasta-0.2.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_google_pasta_sdist_c9f2c8df\":[{\"filename\":\"google-pasta-0.2.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "grpcio": "{\"ncore_pip_deps_311_grpcio_cp311_cp311_linux_armv7l_17325b0b\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-linux_armv7l.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_cp311_cp311_macosx_10_14_universal2_dbe41ad1\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-macosx_10_14_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_cp311_cp311_manylinux_2_17_aarch64_5ea67c72\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-manylinux_2_17_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_cp311_cp311_manylinux_2_17_x86_64_e7831a0f\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_cp311_cp311_musllinux_1_1_x86_64_d6376496\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-musllinux_1_1_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_cp311_cp311_win_amd64_d24035d4\":[{\"filename\":\"grpcio-1.70.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_grpcio_sdist_8d1584a6\":[{\"filename\":\"grpcio-1.70.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "h11": "{\"ncore_pip_deps_311_h11_py3_none_any_63cf8bbe\":[{\"filename\":\"h11-0.16.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_h11_sdist_4e35b956\":[{\"filename\":\"h11-0.16.0.tar.gz\",\"version\":\"3.11\"}]}",
@@ -27786,7 +28418,7 @@
                 "mistune": "{\"ncore_pip_deps_311_mistune_py2_none_any_bad7f5d4\":[{\"filename\":\"mistune-2.0.5-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_mistune_sdist_0246113c\":[{\"filename\":\"mistune-2.0.5.tar.gz\",\"version\":\"3.11\"}]}",
                 "ml_dtypes": "{\"ncore_pip_deps_311_ml_dtypes_cp311_cp311_macosx_10_9_universal2_6c7ecb74\":[{\"filename\":\"ml_dtypes-0.5.4-cp311-cp311-macosx_10_9_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_ml_dtypes_cp311_cp311_manylinux_2_27_aarch64_bc11d7e8\":[{\"filename\":\"ml_dtypes-0.5.4-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_ml_dtypes_cp311_cp311_manylinux_2_27_x86_64_19b9a535\":[{\"filename\":\"ml_dtypes-0.5.4-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_ml_dtypes_cp311_cp311_win_amd64_7c23c54a\":[{\"filename\":\"ml_dtypes-0.5.4-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_ml_dtypes_sdist_8ab06a50\":[{\"filename\":\"ml_dtypes-0.5.4.tar.gz\",\"version\":\"3.11\"}]}",
                 "mpmath": "{\"ncore_pip_deps_311_mpmath_py3_none_any_a0b2b9fe\":[{\"filename\":\"mpmath-1.3.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_mpmath_sdist_7a28eb2a\":[{\"filename\":\"mpmath-1.3.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_mpmath_py3_none_any_a0b2b9fe\":[{\"filename\":\"mpmath-1.3.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_mpmath_sdist_7a28eb2a\":[{\"filename\":\"mpmath-1.3.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_mpmath_py3_none_any_a0b2b9fe\":[{\"filename\":\"mpmath-1.3.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_mpmath_sdist_7a28eb2a\":[{\"filename\":\"mpmath-1.3.0.tar.gz\",\"version\":\"3.13\"}]}",
-                "msgspec": "{\"ncore_pip_deps_311_msgspec_cp311_cp311_macosx_10_9_x86_64_09e0efbf\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_macosx_11_0_arm64_23ee3787\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_manylinux_2_17_aarch64_81f4ac6f\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_manylinux_2_17_x86_64_bb4d873f\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_musllinux_1_2_aarch64_b92b8334\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_musllinux_1_2_x86_64_562c44b0\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_win_amd64_d1dcc93a\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_sdist_692349e5\":[{\"filename\":\"msgspec-0.20.0.tar.gz\",\"version\":\"3.11\"}]}",
+                "msgspec": "{\"ncore_pip_deps_311_msgspec_cp311_cp311_macosx_10_9_x86_64_09e0efbf\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_macosx_11_0_arm64_23ee3787\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_manylinux_2_17_aarch64_81f4ac6f\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_manylinux_2_17_x86_64_bb4d873f\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_musllinux_1_2_aarch64_b92b8334\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_musllinux_1_2_x86_64_562c44b0\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_cp311_cp311_win_amd64_d1dcc93a\":[{\"filename\":\"msgspec-0.20.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_msgspec_sdist_692349e5\":[{\"filename\":\"msgspec-0.20.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_macosx_10_13_x86_64_f13c127a\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_macosx_11_0_arm64_5aa24eb4\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_manylinux_2_17_aarch64_627bfdfe\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_manylinux_2_17_x86_64_c6c310ef\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_musllinux_1_2_aarch64_bc374ded\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-musllinux_1_2_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_musllinux_1_2_x86_64_6f48317f\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-musllinux_1_2_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_cp313_cp313_win_amd64_0739b068\":[{\"filename\":\"msgspec-0.22.0-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_msgspec_sdist_0a13624a\":[{\"filename\":\"msgspec-0.22.0.tar.gz\",\"version\":\"3.13\"}]}",
                 "mypy_extensions": "{\"ncore_pip_deps_311_mypy_extensions_py3_none_any_4392f6c0\":[{\"filename\":\"mypy_extensions-1.0.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_mypy_extensions_sdist_75dbf895\":[{\"filename\":\"mypy_extensions-1.0.0.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_mypy_extensions_py3_none_any_1be4cccd\":[{\"filename\":\"mypy_extensions-1.1.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_mypy_extensions_sdist_52e68efc\":[{\"filename\":\"mypy_extensions-1.1.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_mypy_extensions_py3_none_any_1be4cccd\":[{\"filename\":\"mypy_extensions-1.1.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_mypy_extensions_sdist_52e68efc\":[{\"filename\":\"mypy_extensions-1.1.0.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_mypy_extensions_py3_none_any_4392f6c0\":[{\"filename\":\"mypy_extensions-1.0.0-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_mypy_extensions_sdist_75dbf895\":[{\"filename\":\"mypy_extensions-1.0.0.tar.gz\",\"version\":\"3.8\"}]}",
                 "namex": "{\"ncore_pip_deps_311_namex_py3_none_any_7ddb6c2b\":[{\"filename\":\"namex-0.0.8-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_namex_sdist_32a50f6c\":[{\"filename\":\"namex-0.0.8.tar.gz\",\"version\":\"3.11\"}]}",
                 "nbclient": "{\"ncore_pip_deps_311_nbclient_py3_none_any_4ffee11e\":[{\"filename\":\"nbclient-0.10.2-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_nbclient_sdist_90b7fc6b\":[{\"filename\":\"nbclient-0.10.2.tar.gz\",\"version\":\"3.11\"}]}",
@@ -27794,7 +28426,7 @@
                 "nbformat": "{\"ncore_pip_deps_311_nbformat_py3_none_any_3b48d6c8\":[{\"filename\":\"nbformat-5.10.4-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_nbformat_sdist_322168b1\":[{\"filename\":\"nbformat-5.10.4.tar.gz\",\"version\":\"3.11\"}]}",
                 "nbsphinx": "{\"ncore_pip_deps_311_nbsphinx_py3_none_any_6e805e96\":[{\"filename\":\"nbsphinx-0.9.3-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_nbsphinx_sdist_ec339c86\":[{\"filename\":\"nbsphinx-0.9.3.tar.gz\",\"version\":\"3.11\"}]}",
                 "networkx": "{\"ncore_pip_deps_311_networkx_py3_none_any_d47fbf30\":[{\"filename\":\"networkx-3.6.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_networkx_sdist_26b7c357\":[{\"filename\":\"networkx-3.6.1.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_networkx_py3_none_any_e3fd2c13\":[{\"filename\":\"networkx-3.7-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_networkx_sdist_fd77a511\":[{\"filename\":\"networkx-3.7.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_networkx_py3_none_any_e3fd2c13\":[{\"filename\":\"networkx-3.7-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_networkx_sdist_fd77a511\":[{\"filename\":\"networkx-3.7.tar.gz\",\"version\":\"3.13\"}]}",
-                "numcodecs": "{\"ncore_pip_deps_311_numcodecs_cp311_cp311_macosx_10_9_x86_64_698f1d59\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_macosx_11_0_arm64_bef8c8e6\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_manylinux_2_17_x86_64_cdfaef9f\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_win_amd64_e2547fa3\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_sdist_eeed77e4\":[{\"filename\":\"numcodecs-0.15.1.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_macosx_10_13_x86_64_b0a9d9cd\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-macosx_10_13_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_macosx_11_0_arm64_a34f0fe5\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-macosx_11_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_manylinux_2_17_x86_64_c3a09e22\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_win_amd64_daed6066\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-win_amd64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_sdist_eeed77e4\":[{\"filename\":\"numcodecs-0.15.1.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_10_13_x86_64_e3d82b70\":[{\"filename\":\"numcodecs-0.15.1-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_11_0_arm64_1d471a18\":[{\"filename\":\"numcodecs-0.15.1-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_x86_64_1dfdea4a\":[{\"filename\":\"numcodecs-0.15.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_win_amd64_a4f7bdb2\":[{\"filename\":\"numcodecs-0.15.1-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_sdist_eeed77e4\":[{\"filename\":\"numcodecs-0.15.1.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_macosx_10_9_x86_64_0c240858\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-macosx_10_9_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_manylinux_2_17_x86_64_ee5bda16\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_win_amd64_bd05cdb8\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-win_amd64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_sdist_6c058b32\":[{\"filename\":\"numcodecs-0.11.0.tar.gz\",\"version\":\"3.8\"}]}",
+                "numcodecs": "{\"ncore_pip_deps_311_numcodecs_cp311_cp311_macosx_10_9_x86_64_698f1d59\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_macosx_11_0_arm64_bef8c8e6\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_manylinux_2_17_x86_64_cdfaef9f\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_cp311_cp311_win_amd64_e2547fa3\":[{\"filename\":\"numcodecs-0.15.1-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numcodecs_sdist_eeed77e4\":[{\"filename\":\"numcodecs-0.15.1.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_macosx_10_13_x86_64_b0a9d9cd\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-macosx_10_13_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_macosx_11_0_arm64_a34f0fe5\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-macosx_11_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_manylinux_2_17_x86_64_c3a09e22\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_cp312_cp312_win_amd64_daed6066\":[{\"filename\":\"numcodecs-0.15.1-cp312-cp312-win_amd64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numcodecs_sdist_eeed77e4\":[{\"filename\":\"numcodecs-0.15.1.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_10_13_x86_64_de2c66db\":[{\"filename\":\"numcodecs-0.17.0-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_macosx_11_0_arm64_69b9b468\":[{\"filename\":\"numcodecs-0.17.0-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_aarch64_7065b334\":[{\"filename\":\"numcodecs-0.17.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_manylinux_2_17_x86_64_6c3342d9\":[{\"filename\":\"numcodecs-0.17.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_cp313_cp313_win_amd64_a854e9c8\":[{\"filename\":\"numcodecs-0.17.0-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numcodecs_sdist_e8db2e33\":[{\"filename\":\"numcodecs-0.17.0.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_macosx_10_9_x86_64_0c240858\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-macosx_10_9_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_manylinux_2_17_x86_64_ee5bda16\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_cp38_cp38_win_amd64_bd05cdb8\":[{\"filename\":\"numcodecs-0.11.0-cp38-cp38-win_amd64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numcodecs_sdist_6c058b32\":[{\"filename\":\"numcodecs-0.11.0.tar.gz\",\"version\":\"3.8\"}]}",
                 "numpy": "{\"ncore_pip_deps_311_numpy_cp311_cp311_macosx_10_9_x86_64_4c66707f\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_macosx_11_0_arm64_edd8b5fe\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_manylinux_2_17_aarch64_7ab55401\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_manylinux_2_17_x86_64_666dbfb6\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_musllinux_1_1_aarch64_96ff0b2a\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-musllinux_1_1_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_musllinux_1_1_x86_64_60dedbb9\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-musllinux_1_1_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_cp311_cp311_win_amd64_cd25bcec\":[{\"filename\":\"numpy-1.26.4-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_numpy_sdist_2a02aba9\":[{\"filename\":\"numpy-1.26.4.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_macosx_10_13_x86_64_c6342f54\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-macosx_10_13_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_macosx_11_0_arm64_b11e8fda\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-macosx_11_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_macosx_14_0_arm64_9cb18a32\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-macosx_14_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_macosx_14_0_x86_64_aec3fc4b\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-macosx_14_0_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_manylinux_2_27_aarch64_fe4d21ab\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_manylinux_2_27_x86_64_fbde6962\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_musllinux_1_2_aarch64_381a7a3d\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-musllinux_1_2_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_musllinux_1_2_x86_64_b89d0aaa\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-musllinux_1_2_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_cp312_cp312_win_amd64_a7b1b635\":[{\"filename\":\"numpy-2.5.4-cp312-cp312-win_amd64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_numpy_sdist_9a94cf75\":[{\"filename\":\"numpy-2.5.4.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_macosx_10_13_x86_64_2377da2d\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_macosx_11_0_arm64_7415db95\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_arm64_6d6a71b9\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-macosx_14_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_macosx_14_0_x86_64_d8200f16\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-macosx_14_0_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_aarch64_1c2e71b0\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_manylinux_2_27_x86_64_6ffa0766\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_aarch64_2fa3328f\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-musllinux_1_2_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_musllinux_1_2_x86_64_b86966fb\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-musllinux_1_2_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_cp313_cp313_win_amd64_8b4d2fd2\":[{\"filename\":\"numpy-2.5.4-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_numpy_sdist_9a94cf75\":[{\"filename\":\"numpy-2.5.4.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_numpy_cp38_cp38_macosx_10_9_x86_64_cf240200\":[{\"filename\":\"numpy-1.19.5-cp38-cp38-macosx_10_9_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numpy_cp38_cp38_manylinux_2_12_x86_64_a9d17f2b\":[{\"filename\":\"numpy-1.19.5-cp38-cp38-manylinux2010_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numpy_cp38_cp38_manylinux_2_17_aarch64_99abf4f3\":[{\"filename\":\"numpy-1.19.5-cp38-cp38-manylinux2014_aarch64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numpy_cp38_cp38_manylinux_2_5_x86_64_012426a4\":[{\"filename\":\"numpy-1.19.5-cp38-cp38-manylinux1_x86_64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numpy_cp38_cp38_win_amd64_811daee3\":[{\"filename\":\"numpy-1.19.5-cp38-cp38-win_amd64.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_numpy_sdist_a76f5024\":[{\"filename\":\"numpy-1.19.5.zip\",\"version\":\"3.8\"}]}",
                 "nuscenes_devkit": "{\"ncore_pip_deps_311_nuscenes_devkit_py3_none_any_76cee0e7\":[{\"filename\":\"nuscenes_devkit-1.2.0-py3-none-any.whl\",\"version\":\"3.11\"}]}",
                 "nvidia_cublas_cu12": "{\"ncore_pip_deps_311_nvidia_cublas_cu12_py3_none_manylinux_2_27_aarch64_93a4e0e3\":[{\"filename\":\"nvidia_cublas_cu12-12.8.3.14-py3-none-manylinux_2_27_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_nvidia_cublas_cu12_py3_none_manylinux_2_27_x86_64_3f0e05e7\":[{\"filename\":\"nvidia_cublas_cu12-12.8.3.14-py3-none-manylinux_2_27_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_nvidia_cublas_cu12_py3_none_win_amd64_9ae5eae5\":[{\"filename\":\"nvidia_cublas_cu12-12.8.3.14-py3-none-win_amd64.whl\",\"version\":\"3.11\"}]}",
@@ -27841,6 +28473,7 @@
                 "pytest": "{\"ncore_pip_deps_311_pytest_py3_none_any_50e16d95\":[{\"filename\":\"pytest-8.3.4-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pytest_sdist_965370d0\":[{\"filename\":\"pytest-8.3.4.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_pytest_py3_none_any_37a86b45\":[{\"filename\":\"pytest-9.1.1-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_pytest_sdist_1088fbde\":[{\"filename\":\"pytest-9.1.1.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_pytest_py3_none_any_37a86b45\":[{\"filename\":\"pytest-9.1.1-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pytest_sdist_1088fbde\":[{\"filename\":\"pytest-9.1.1.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_pytest_py3_none_any_78bf1645\":[{\"filename\":\"pytest-7.4.0-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_pytest_sdist_b4bf8c45\":[{\"filename\":\"pytest-7.4.0.tar.gz\",\"version\":\"3.8\"}]}",
                 "python_dateutil": "{\"ncore_pip_deps_311_python_dateutil_py2_none_any_a8b2bc7b\":[{\"filename\":\"python_dateutil-2.9.0.post0-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_python_dateutil_sdist_37dd5420\":[{\"filename\":\"python-dateutil-2.9.0.post0.tar.gz\",\"version\":\"3.11\"}]}",
                 "pytz": "{\"ncore_pip_deps_311_pytz_py2_none_any_89dd22dc\":[{\"filename\":\"pytz-2025.1-py2.py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pytz_sdist_c2db42be\":[{\"filename\":\"pytz-2025.1.tar.gz\",\"version\":\"3.11\"}]}",
+                "pyyaml": "{\"ncore_pip_deps_311_pyyaml_cp311_cp311_macosx_10_13_x86_64_44edc647\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-macosx_10_13_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_macosx_11_0_arm64_652cb6ed\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_aarch64_10892704\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_s390x_850774a7\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_manylinux_2_17_x86_64_b8bb0864\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_musllinux_1_2_aarch64_1d37d57a\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_musllinux_1_2_x86_64_37503bfb\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_cp311_cp311_win_amd64_9f3bfb49\":[{\"filename\":\"pyyaml-6.0.3-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyyaml_sdist_d7662337\":[{\"filename\":\"pyyaml-6.0.3.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_macosx_10_13_x86_64_8da9669d\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_macosx_11_0_arm64_2283a07e\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_aarch64_ee292290\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_s390x_a33284e2\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_manylinux_2_17_x86_64_0f29edc4\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_musllinux_1_2_aarch64_f7057c9a\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-musllinux_1_2_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_musllinux_1_2_x86_64_eda16858\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-musllinux_1_2_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_cp313_cp313_win_amd64_79005a0d\":[{\"filename\":\"pyyaml-6.0.3-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_pyyaml_sdist_d7662337\":[{\"filename\":\"pyyaml-6.0.3.tar.gz\",\"version\":\"3.13\"}]}",
                 "pyzmq": "{\"ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_15_universal2_c0598838\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-macosx_10_15_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_macosx_10_9_x86_64_ed038a92\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_aarch64_9027a7fc\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_17_x86_64_f0019cc8\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_manylinux_2_28_x86_64_f19dae58\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_musllinux_1_1_aarch64_f5eeeb82\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-musllinux_1_1_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_musllinux_1_1_x86_64_817fcd33\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-musllinux_1_1_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_cp311_cp311_win_amd64_ef29630f\":[{\"filename\":\"pyzmq-26.2.1-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_pyzmq_sdist_17d72a74\":[{\"filename\":\"pyzmq-26.2.1.tar.gz\",\"version\":\"3.11\"}]}",
                 "referencing": "{\"ncore_pip_deps_311_referencing_py3_none_any_d53ae300\":[{\"filename\":\"referencing-0.34.0-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_referencing_sdist_5773bd84\":[{\"filename\":\"referencing-0.34.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "remotezip": "{\"ncore_pip_deps_311_remotezip_py3_none_any_f70a4026\":[{\"filename\":\"remotezip-0.12.3-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_remotezip_sdist_bf1ebe2b\":[{\"filename\":\"remotezip-0.12.3.tar.gz\",\"version\":\"3.11\"}]}",
@@ -27892,10 +28525,10 @@
                 "websockets": "{\"ncore_pip_deps_311_websockets_cp311_cp311_macosx_10_9_universal2_823c248b\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-macosx_10_9_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_macosx_10_9_x86_64_67899970\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_macosx_11_0_arm64_d50fd1ee\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_manylinux_2_17_aarch64_d99e5546\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_manylinux_2_5_x86_64_8dd8327c\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_musllinux_1_2_aarch64_8fdc5105\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_musllinux_1_2_x86_64_54479983\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_cp311_cp311_win_amd64_27ccee00\":[{\"filename\":\"websockets-15.0.1-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_py3_none_any_f7a866fb\":[{\"filename\":\"websockets-15.0.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_websockets_sdist_82544de0\":[{\"filename\":\"websockets-15.0.1.tar.gz\",\"version\":\"3.11\"}]}",
                 "werkzeug": "{\"ncore_pip_deps_311_werkzeug_py3_none_any_63a77fb8\":[{\"filename\":\"werkzeug-3.1.8-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_werkzeug_sdist_9bad61a4\":[{\"filename\":\"werkzeug-3.1.8.tar.gz\",\"version\":\"3.11\"}]}",
                 "wheel": "{\"ncore_pip_deps_311_wheel_py3_none_any_708e7481\":[{\"filename\":\"wheel-0.45.1-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wheel_sdist_661e1abd\":[{\"filename\":\"wheel-0.45.1.tar.gz\",\"version\":\"3.11\"}]}",
-                "wrapt": "{\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_10_9_universal2_ff04ef6e\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_10_9_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_10_9_x86_64_4db983e7\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_11_0_arm64_9abc77a4\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_manylinux_2_17_aarch64_0b929ac1\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_manylinux_2_5_x86_64_1a7ed2d9\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_musllinux_1_2_aarch64_129a150f\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_musllinux_1_2_x86_64_9a2bce78\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_win_amd64_acc130bc\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_py3_none_any_b18f2d15\":[{\"filename\":\"wrapt-1.17.2-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_sdist_41388e9d\":[{\"filename\":\"wrapt-1.17.2.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_macosx_10_13_x86_64_b898caea\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-macosx_10_13_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_macosx_11_0_arm64_8837fbe7\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-macosx_11_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_manylinux_2_17_aarch64_6761765c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_manylinux_2_5_x86_64_0cabb9c1\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_musllinux_1_2_aarch64_6a9ee62a\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-musllinux_1_2_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_musllinux_1_2_x86_64_e716f47c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-musllinux_1_2_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_win_amd64_ab45839c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-win_amd64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_py3_none_any_107eea1a\":[{\"filename\":\"wrapt-2.5.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_sdist_c48cdb6c\":[{\"filename\":\"wrapt-2.5.0.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_macosx_10_13_x86_64_b312b3cc\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-macosx_10_13_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_macosx_11_0_arm64_c57ddae2\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-macosx_11_0_arm64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_manylinux_2_17_aarch64_6058e12e\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_manylinux_2_5_x86_64_b95a6eca\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_musllinux_1_2_aarch64_f4ef4935\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-musllinux_1_2_aarch64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_musllinux_1_2_x86_64_a346408f\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-musllinux_1_2_x86_64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_cp313_cp313_win_amd64_77f0a74f\":[{\"filename\":\"wrapt-2.5.0-cp313-cp313-win_amd64.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_py3_none_any_107eea1a\":[{\"filename\":\"wrapt-2.5.0-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_wrapt_sdist_c48cdb6c\":[{\"filename\":\"wrapt-2.5.0.tar.gz\",\"version\":\"3.13\"}]}",
+                "wrapt": "{\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_10_9_universal2_ff04ef6e\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_10_9_universal2.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_10_9_x86_64_4db983e7\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_macosx_11_0_arm64_9abc77a4\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_manylinux_2_17_aarch64_0b929ac1\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_manylinux_2_5_x86_64_1a7ed2d9\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_musllinux_1_2_aarch64_129a150f\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_musllinux_1_2_x86_64_9a2bce78\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_cp311_cp311_win_amd64_acc130bc\":[{\"filename\":\"wrapt-1.17.2-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_py3_none_any_b18f2d15\":[{\"filename\":\"wrapt-1.17.2-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_wrapt_sdist_41388e9d\":[{\"filename\":\"wrapt-1.17.2.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_macosx_10_13_x86_64_b898caea\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-macosx_10_13_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_macosx_11_0_arm64_8837fbe7\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-macosx_11_0_arm64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_manylinux_2_17_aarch64_6761765c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_manylinux_2_5_x86_64_0cabb9c1\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_musllinux_1_2_aarch64_6a9ee62a\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-musllinux_1_2_aarch64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_musllinux_1_2_x86_64_e716f47c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-musllinux_1_2_x86_64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_cp312_cp312_win_amd64_ab45839c\":[{\"filename\":\"wrapt-2.5.0-cp312-cp312-win_amd64.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_py3_none_any_107eea1a\":[{\"filename\":\"wrapt-2.5.0-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_wrapt_sdist_c48cdb6c\":[{\"filename\":\"wrapt-2.5.0.tar.gz\",\"version\":\"3.12\"}]}",
                 "xxhash": "{\"ncore_pip_deps_311_xxhash_cp311_cp311_macosx_10_9_x86_64_b47bbd8c\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_macosx_11_0_arm64_2b6821e9\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_manylinux_2_17_aarch64_8b29ee68\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_manylinux_2_17_ppc64le_6812c25f\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_manylinux_2_17_s390x_4ccbff01\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_manylinux_2_17_x86_64_297b7fbf\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_musllinux_1_2_aarch64_dea26ae1\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_musllinux_1_2_ppc64le_08d45aef\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-musllinux_1_2_ppc64le.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_musllinux_1_2_s390x_92914236\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-musllinux_1_2_s390x.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_musllinux_1_2_x86_64_51312c76\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_cp311_cp311_win_amd64_26734cdc\":[{\"filename\":\"xxhash-3.6.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_xxhash_sdist_f0162a78\":[{\"filename\":\"xxhash-3.6.0.tar.gz\",\"version\":\"3.11\"}]}",
                 "yourdfpy": "{\"ncore_pip_deps_311_yourdfpy_py3_none_any_8a187a8b\":[{\"filename\":\"yourdfpy-0.0.60-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_yourdfpy_sdist_2af2d8bd\":[{\"filename\":\"yourdfpy-0.0.60.tar.gz\",\"version\":\"3.11\"}]}",
-                "zarr": "{\"ncore_pip_deps_311_zarr_py3_none_any_2795e20a\":[{\"filename\":\"zarr-2.18.4-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zarr_sdist_37790ede\":[{\"filename\":\"zarr-2.18.4.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_zarr_py3_none_any_ac3dc403\":[{\"filename\":\"zarr-2.18.7-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_zarr_sdist_b2b8f66f\":[{\"filename\":\"zarr-2.18.7.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_zarr_py3_none_any_ac3dc403\":[{\"filename\":\"zarr-2.18.7-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_zarr_sdist_b2b8f66f\":[{\"filename\":\"zarr-2.18.7.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_zarr_py3_none_any_883305e8\":[{\"filename\":\"zarr-2.13.3-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_zarr_sdist_db24b090\":[{\"filename\":\"zarr-2.13.3.tar.gz\",\"version\":\"3.8\"}]}",
+                "zarr": "{\"ncore_pip_deps_311_zarr_py3_none_any_72c3dcee\":[{\"filename\":\"zarr-3.1.4-py3-none-any.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zarr_sdist_b0cf4eea\":[{\"filename\":\"zarr-3.1.4.tar.gz\",\"version\":\"3.11\"}],\"ncore_pip_deps_312_zarr_py3_none_any_ac3dc403\":[{\"filename\":\"zarr-2.18.7-py3-none-any.whl\",\"version\":\"3.12\"}],\"ncore_pip_deps_312_zarr_sdist_b2b8f66f\":[{\"filename\":\"zarr-2.18.7.tar.gz\",\"version\":\"3.12\"}],\"ncore_pip_deps_313_zarr_py3_none_any_38b54057\":[{\"filename\":\"zarr-3.4.1-py3-none-any.whl\",\"version\":\"3.13\"}],\"ncore_pip_deps_313_zarr_sdist_b34bda11\":[{\"filename\":\"zarr-3.4.1.tar.gz\",\"version\":\"3.13\"}],\"ncore_pip_deps_38_zarr_py3_none_any_883305e8\":[{\"filename\":\"zarr-2.13.3-py3-none-any.whl\",\"version\":\"3.8\"}],\"ncore_pip_deps_38_zarr_sdist_db24b090\":[{\"filename\":\"zarr-2.13.3.tar.gz\",\"version\":\"3.8\"}]}",
                 "zstandard": "{\"ncore_pip_deps_311_zstandard_cp311_cp311_macosx_10_9_x86_64_933b65d7\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-macosx_10_9_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_macosx_11_0_arm64_a3f79487\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-macosx_11_0_arm64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_manylinux_2_17_aarch64_01582723\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_manylinux_2_17_ppc64le_5f1ad7bf\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_manylinux_2_17_s390x_06acb75e\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_manylinux_2_17_x86_64_9300d02e\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_1_aarch64_bfd06b1c\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_1_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_1_x86_64_f373da2c\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_1_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_2_aarch64_6c0e5a65\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_2_aarch64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_2_ppc64le_98750a30\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_2_ppc64le.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_2_s390x_22a086cf\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_2_s390x.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_musllinux_1_2_x86_64_72d35d7a\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-musllinux_1_2_x86_64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_cp311_cp311_win_amd64_daab68fa\":[{\"filename\":\"zstandard-0.25.0-cp311-cp311-win_amd64.whl\",\"version\":\"3.11\"}],\"ncore_pip_deps_311_zstandard_sdist_7713e117\":[{\"filename\":\"zstandard-0.25.0.tar.gz\",\"version\":\"3.11\"}]}"
               },
               "packages": [
@@ -27925,6 +28558,7 @@
                 "deprecated",
                 "descartes",
                 "docutils",
+                "donfig",
                 "dracopy",
                 "embreex",
                 "entrypoints",
@@ -27938,6 +28572,7 @@
                 "fonttools",
                 "fsspec",
                 "gast",
+                "google_crc32c",
                 "google_pasta",
                 "grpcio",
                 "h11",
@@ -28029,6 +28664,7 @@
                 "pytest",
                 "python_dateutil",
                 "pytz",
+                "pyyaml",
                 "pyzmq",
                 "referencing",
                 "remotezip",
diff --git a/deps/pip/requirements_3_11.in b/deps/pip/requirements_3_11.in
index b31bf737..ba8ade94 100644
--- a/deps/pip/requirements_3_11.in
+++ b/deps/pip/requirements_3_11.in
@@ -26,6 +26,7 @@
 
 # Public API restrictions for 3.11
 # torch 2.7+ supports numpy 2.x natively, no constraint needed
+zarr==3.1.4 # oldest supported zarr-python 3 (most recent zarr-python 3 is covered by python 3.13)
 
 --find-links https://download.pytorch.org/whl/cu128/torch
 torch==2.7.0+cu128
diff --git a/deps/pip/requirements_3_11.txt b/deps/pip/requirements_3_11.txt
index d620e757..19e1ac35 100644
--- a/deps/pip/requirements_3_11.txt
+++ b/deps/pip/requirements_3_11.txt
@@ -22,9 +22,6 @@ anyio==4.12.1 \
     --hash=sha256:41cfcc3a4c85d3f05c932da7c26d0201ac36f72abd4435ba90d0464a3ffed703 \
     --hash=sha256:d405828884fc140aa80a3c667b8beed277f1dfedec42ba031bd6ac3db606ab6c
     # via httpx
-asciitree==0.3.3 \
-    --hash=sha256:4aa4b9b649f85e3fcb343363d97564aa1fb62e249677f2e18a96765145cc0f6e
-    # via zarr
 asttokens==3.0.1 \
     --hash=sha256:15a3ebc0f43c2d0a50eeafea25e19046c68398e487b9f1f5b517f7c0f40f976a \
     --hash=sha256:71a4ee5de0bde6a31d64f6b13f2293ac190344478f081c3d1bccfcf5eacb0cb7
@@ -337,6 +334,10 @@ docutils==0.22.4 \
     #   nbsphinx
     #   pydata-sphinx-theme
     #   sphinx
+donfig==0.8.1.post1 \
+    --hash=sha256:2a3175ce74a06109ff9307d90a230f81215cbac9a751f4d1c6194644b8204f9d \
+    --hash=sha256:3bef3413a4c1c601b585e8d297256d0c1470ea012afa6e8461dc28bfb7c23f52
+    # via zarr
 dracopy==2.0.0 \
     --hash=sha256:00ed24e138bcfbf2605d495c3a4e6c17b71cbd0b63b78e945d9d3c89d4fb9d69 \
     --hash=sha256:05b5d4054402877f50221bc917e2943214a956d8d746e01b21f57ecc2d0adf21 \
@@ -416,10 +417,6 @@ executing==2.2.0 \
     --hash=sha256:11387150cad388d62750327a53d3339fad4888b39a6fe233c3afbb54ecffd3aa \
     --hash=sha256:5d108c028108fe2551d1a7b2e8b713341e2cb4fc0aa7dcf966fa4327a5226755
     # via stack-data
-fasteners==0.19 \
-    --hash=sha256:758819cb5d94cdedf4e836988b74de396ceacb8e2794d21f82d131fd9ee77237 \
-    --hash=sha256:b4f37c3ac52d8a445af3a66bce57b33b5e90b97c696b7b984f530cf8f0ded09c
-    # via zarr
 fastjsonschema==2.21.1 \
     --hash=sha256:794d4f0a58f848961ba16af7b9c85a3e88cd360df008c59aac6fc5ae9323b5d4 \
     --hash=sha256:c9e5b7e908310918cf494a434eeb31384dd84a98b57a30bcb1f535015b554667
@@ -497,6 +494,39 @@ gast==0.6.0 \
     --hash=sha256:52b182313f7330389f72b069ba00f174cfe2a06411099547288839c6cbafbd54 \
     --hash=sha256:88fc5300d32c7ac6ca7b515310862f71e6fdf2c029bbec7c66c0f5dd47b6b1fb
     # via tensorflow
+google-crc32c==1.9.0 \
+    --hash=sha256:0568b17ed90ac596f29400d99e243fd0cc6276766183def888d1bf8d1dc13827 \
+    --hash=sha256:0abe7e202c25909869c35672ab0f2fe748a7acf276eb78577332a7c38999740f \
+    --hash=sha256:13179f7e3282617923e957b8e54b8f9c3968030f48640a9f47fd7c5c38c4a215 \
+    --hash=sha256:16865b477d7941712cb0e0aad8ad4815e984fb5fc16d3fdaef7d986e26e53c95 \
+    --hash=sha256:265233aff33d835f5b909584fe36ab29647b598c271b661a300001099109e53e \
+    --hash=sha256:280f3a3e47af0eeba3a3e5aa7d311af77001812b8df80fb8beafcd0b40eaf7f1 \
+    --hash=sha256:3abb18297d9ef0ab120531838be0e6d68c9fa876570e11c229c48f2edac23ce7 \
+    --hash=sha256:43a2dc26f9be213fbe0b4fc4a1088c5d45cbfcb3247420ccc820f0fc3edeea86 \
+    --hash=sha256:4488f1553a9ab7e86cdedc833374a7e904031803b995dc0bd0be48c271fa6556 \
+    --hash=sha256:457d0d9a4718fd52b1494eac5c200ad25beeadbdc91843d550a003910838589f \
+    --hash=sha256:51cb4e23a38ad4f495f35f87c233ca3ea6b9c4559e7ac383cdef786fab0f7977 \
+    --hash=sha256:53fdafef58e230d0c946ab5f8446d123d9f548230a73b29c8b41c9546f268bc1 \
+    --hash=sha256:56610f548f1b35c9568b9d1de30423480f505dae4991556072d5802820ff35c4 \
+    --hash=sha256:5695c8b9327e040b2aba12c6659b0acb5995314ef0af0192da66e662e011103b \
+    --hash=sha256:6a3b2c8a343c570ed8100a7627c20badfd92c6caa2067093a86be45af27f5b1b \
+    --hash=sha256:7b8c84c3d159ab6817fe3f74e6e6cef099c3f95dcec3abc0d8afb1404642efbe \
+    --hash=sha256:8535e75dfead304f30e9122b9ea2c0a570dbaa52c176a0a591540c7914c1e46d \
+    --hash=sha256:8583ec21d56b565d68ab2963cc7e21b3b271247c29b04286068255ef65f221bd \
+    --hash=sha256:86764b99e7a607830d93cb5b75e0ec3ff6cb06d3c274624418473cee701900d4 \
+    --hash=sha256:8b91f41645b15a720357183fa5716682ada441873e3c462c15f9714be36f146b \
+    --hash=sha256:af73200fa9791ccd380f3598235dba8d82b8af0905df045b3dc60b59836e8ddd \
+    --hash=sha256:ccfe40021fd6afe23361175cf7551e3cef5fd34dc1ebe319f14993a83579e0eb \
+    --hash=sha256:dee799544cae42a42b17a88e38b59cf2c271051dc001da2117a8ff240ffa0548 \
+    --hash=sha256:e6b529a6a287104ec79d281c411685231200ce954a29c28ab8e5093cb6e130fb \
+    --hash=sha256:e6e8be8a94436079cb5340f6d495d9d7ba30124d8b952703994c739c7c06e236 \
+    --hash=sha256:f1dc17d987ddcc5eba12a7ce48f0eb93141dea236b170c1101151396edf2f0cf \
+    --hash=sha256:f2b64641bca27497b986b9d87883014035aa904cb4fa333407c6752b3afee9ba \
+    --hash=sha256:f894a2877650b56201d26a012a257b76d54a68834dc3913a93830ca8a047b075 \
+    --hash=sha256:f97c3806dcea41c29c04965347b0e12481561b75e0045dc7a4f69d75dec5d9b1 \
+    --hash=sha256:fb63a8d7fa2e95dcff1ca16af2f4d88b526fa5ff72d1696285884ac2d49b6963 \
+    --hash=sha256:fbef61a3794e011c65fb4396a196cf123a7f474fe5a443db8e5dd7d751b9e6d4
+    # via zarr
 google-pasta==0.2.0 \
     --hash=sha256:4612951da876b1a10fe3960d7226f0c7682cf901e16ac06e473b267a5afa8954 \
     --hash=sha256:b32482794a366b5366a32c92a9a9201b107821889935a02b3e51f6b432ea84ed \
@@ -1563,6 +1593,7 @@ packaging==24.2 \
     #   sphinx
     #   tensorboard
     #   tensorflow
+    #   zarr
 pandas==2.2.3 \
     --hash=sha256:062309c1b9ea12a50e8ce661145c6aab431b1e99530d3cd60640e255778bd43a \
     --hash=sha256:15c0e1e02e93116177d29ff83e8b1619c93ddc9c49083f237d4312337a61165d \
@@ -1960,6 +1991,81 @@ pytz==2025.1 \
     --hash=sha256:89dd22dca55b46eac6eda23b2d72721bf1bdfef212645d81513ef5d03038de57 \
     --hash=sha256:c2db42be2a2518b28e65f9207c4d05e6ff547d1efa4086469ef855e4ab70178e
     # via pandas
+pyyaml==6.0.3 \
+    --hash=sha256:00c4bdeba853cc34e7dd471f16b4114f4162dc03e6b7afcc2128711f0eca823c \
+    --hash=sha256:0150219816b6a1fa26fb4699fb7daa9caf09eb1999f3b70fb6e786805e80375a \
+    --hash=sha256:02893d100e99e03eda1c8fd5c441d8c60103fd175728e23e431db1b589cf5ab3 \
+    --hash=sha256:02ea2dfa234451bbb8772601d7b8e426c2bfa197136796224e50e35a78777956 \
+    --hash=sha256:0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6 \
+    --hash=sha256:10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c \
+    --hash=sha256:16249ee61e95f858e83976573de0f5b2893b3677ba71c9dd36b9cf8be9ac6d65 \
+    --hash=sha256:1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a \
+    --hash=sha256:1ebe39cb5fc479422b83de611d14e2c0d3bb2a18bbcb01f229ab3cfbd8fee7a0 \
+    --hash=sha256:214ed4befebe12df36bcc8bc2b64b396ca31be9304b8f59e25c11cf94a4c033b \
+    --hash=sha256:2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1 \
+    --hash=sha256:22ba7cfcad58ef3ecddc7ed1db3409af68d023b7f940da23c6c2a1890976eda6 \
+    --hash=sha256:27c0abcb4a5dac13684a37f76e701e054692a9b2d3064b70f5e4eb54810553d7 \
+    --hash=sha256:28c8d926f98f432f88adc23edf2e6d4921ac26fb084b028c733d01868d19007e \
+    --hash=sha256:2e71d11abed7344e42a8849600193d15b6def118602c4c176f748e4583246007 \
+    --hash=sha256:34d5fcd24b8445fadc33f9cf348c1047101756fd760b4dacb5c3e99755703310 \
+    --hash=sha256:37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4 \
+    --hash=sha256:3c5677e12444c15717b902a5798264fa7909e41153cdf9ef7ad571b704a63dd9 \
+    --hash=sha256:3ff07ec89bae51176c0549bc4c63aa6202991da2d9a6129d7aef7f1407d3f295 \
+    --hash=sha256:41715c910c881bc081f1e8872880d3c650acf13dfa8214bad49ed4cede7c34ea \
+    --hash=sha256:418cf3f2111bc80e0933b2cd8cd04f286338bb88bdc7bc8e6dd775ebde60b5e0 \
+    --hash=sha256:44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e \
+    --hash=sha256:4a2e8cebe2ff6ab7d1050ecd59c25d4c8bd7e6f400f5f82b96557ac0abafd0ac \
+    --hash=sha256:4ad1906908f2f5ae4e5a8ddfce73c320c2a1429ec52eafd27138b7f1cbe341c9 \
+    --hash=sha256:501a031947e3a9025ed4405a168e6ef5ae3126c59f90ce0cd6f2bfc477be31b7 \
+    --hash=sha256:5190d403f121660ce8d1d2c1bb2ef1bd05b5f68533fc5c2ea899bd15f4399b35 \
+    --hash=sha256:5498cd1645aa724a7c71c8f378eb29ebe23da2fc0d7a08071d89469bf1d2defb \
+    --hash=sha256:5cf4e27da7e3fbed4d6c3d8e797387aaad68102272f8f9752883bc32d61cb87b \
+    --hash=sha256:5e0b74767e5f8c593e8c9b5912019159ed0533c70051e9cce3e8b6aa699fcd69 \
+    --hash=sha256:5ed875a24292240029e4483f9d4a4b8a1ae08843b9c54f43fcc11e404532a8a5 \
+    --hash=sha256:5fcd34e47f6e0b794d17de1b4ff496c00986e1c83f7ab2fb8fcfe9616ff7477b \
+    --hash=sha256:5fdec68f91a0c6739b380c83b951e2c72ac0197ace422360e6d5a959d8d97b2c \
+    --hash=sha256:6344df0d5755a2c9a276d4473ae6b90647e216ab4757f8426893b5dd2ac3f369 \
+    --hash=sha256:64386e5e707d03a7e172c0701abfb7e10f0fb753ee1d773128192742712a98fd \
+    --hash=sha256:652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824 \
+    --hash=sha256:66291b10affd76d76f54fad28e22e51719ef9ba22b29e1d7d03d6777a9174198 \
+    --hash=sha256:66e1674c3ef6f541c35191caae2d429b967b99e02040f5ba928632d9a7f0f065 \
+    --hash=sha256:6adc77889b628398debc7b65c073bcb99c4a0237b248cacaf3fe8a557563ef6c \
+    --hash=sha256:79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c \
+    --hash=sha256:7c6610def4f163542a622a73fb39f534f8c101d690126992300bf3207eab9764 \
+    --hash=sha256:7f047e29dcae44602496db43be01ad42fc6f1cc0d8cd6c83d342306c32270196 \
+    --hash=sha256:8098f252adfa6c80ab48096053f512f2321f0b998f98150cea9bd23d83e1467b \
+    --hash=sha256:850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00 \
+    --hash=sha256:8d1fab6bb153a416f9aeb4b8763bc0f22a5586065f86f7664fc23339fc1c1fac \
+    --hash=sha256:8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8 \
+    --hash=sha256:8dc52c23056b9ddd46818a57b78404882310fb473d63f17b07d5c40421e47f8e \
+    --hash=sha256:9149cad251584d5fb4981be1ecde53a1ca46c891a79788c0df828d2f166bda28 \
+    --hash=sha256:93dda82c9c22deb0a405ea4dc5f2d0cda384168e466364dec6255b293923b2f3 \
+    --hash=sha256:96b533f0e99f6579b3d4d4995707cf36df9100d67e0c8303a0c55b27b5f99bc5 \
+    --hash=sha256:9c57bb8c96f6d1808c030b1687b9b5fb476abaa47f0db9c0101f5e9f394e97f4 \
+    --hash=sha256:9c7708761fccb9397fe64bbc0395abcae8c4bf7b0eac081e12b809bf47700d0b \
+    --hash=sha256:9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf \
+    --hash=sha256:a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5 \
+    --hash=sha256:a80cb027f6b349846a3bf6d73b5e95e782175e52f22108cfa17876aaeff93702 \
+    --hash=sha256:b30236e45cf30d2b8e7b3e85881719e98507abed1011bf463a8fa23e9c3e98a8 \
+    --hash=sha256:b3bc83488de33889877a0f2543ade9f70c67d66d9ebb4ac959502e12de895788 \
+    --hash=sha256:b865addae83924361678b652338317d1bd7e79b1f4596f96b96c77a5a34b34da \
+    --hash=sha256:b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d \
+    --hash=sha256:ba1cc08a7ccde2d2ec775841541641e4548226580ab850948cbfda66a1befcdc \
+    --hash=sha256:bdb2c67c6c1390b63c6ff89f210c8fd09d9a1217a465701eac7316313c915e4c \
+    --hash=sha256:c1ff362665ae507275af2853520967820d9124984e0f7466736aea23d8611fba \
+    --hash=sha256:c2514fceb77bc5e7a2f7adfaa1feb2fb311607c9cb518dbc378688ec73d8292f \
+    --hash=sha256:c3355370a2c156cffb25e876646f149d5d68f5e0a3ce86a5084dd0b64a994917 \
+    --hash=sha256:c458b6d084f9b935061bc36216e8a69a7e293a2f1e68bf956dcd9e6cbcd143f5 \
+    --hash=sha256:d0eae10f8159e8fdad514efdc92d74fd8d682c933a6dd088030f3834bc8e6b26 \
+    --hash=sha256:d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f \
+    --hash=sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b \
+    --hash=sha256:eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be \
+    --hash=sha256:ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c \
+    --hash=sha256:efd7b85f94a6f21e4932043973a7ba2613b059c4a000551892ac9f1d11f5baf3 \
+    --hash=sha256:f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6 \
+    --hash=sha256:fa160448684b4e94d80416c0fa4aac48967a969efe22931448d853ada8baf926 \
+    --hash=sha256:fc09d0aa354569bc501d4e787133afc08552722d3ab34836a80547331bb5d4a0
+    # via donfig
 pyzmq==26.2.1 \
     --hash=sha256:000760e374d6f9d1a3478a42ed0c98604de68c9e94507e5452951e598ebecfba \
     --hash=sha256:004837cb958988c75d8042f5dac19a881f3d9b3b75b2f574055e22573745f841 \
@@ -2545,6 +2651,7 @@ typing-extensions==4.12.2 \
     #   torch
     #   typing-inspect
     #   viser
+    #   zarr
 typing-inspect==0.9.0 \
     --hash=sha256:9ee6fc59062311ef8547596ab6b955e1b8aa46242d854bfc78f4f6b0eff35f9f \
     --hash=sha256:b23fc42ff6f6ef6954e4852c1fb512cdd18dbea03134f91f856a95ccc9461f78
@@ -2946,10 +3053,12 @@ yourdfpy==0.0.60 \
     --hash=sha256:2af2d8bdeea1b85b642590a3b4236fdb35746d7b3e38ce460a169c18d9c4f868 \
     --hash=sha256:8a187a8b18c98db87c76e9a950581b3c875b761e00df83942526c17ea693166c
     # via viser
-zarr==2.18.4 \
-    --hash=sha256:2795e20aff91093ce7e4da36ab1a138aededbd8ab66bf01fd01512e61d31e5d1 \
-    --hash=sha256:37790ededd0683ae1abe6ff90aa16c22543b3436810060f53d72c15e910c24bb
-    # via -r deps/pip/requirements_ncore.in
+zarr==3.1.4 \
+    --hash=sha256:72c3dcee2d41527d21cd814cbaca1e6e23550701eb4fc30767ee53d925900eec \
+    --hash=sha256:b0cf4eea8367ae444d7d0ea90b98d29ac0b7eb691fac2e0bcc2d2bd659522903
+    # via
+    #   -r deps/pip/requirements_3_11.in
+    #   -r deps/pip/requirements_ncore.in
 zstandard==0.25.0 \
     --hash=sha256:011d388c76b11a0c165374ce660ce2c8efa8e5d87f34996aa80f9c0816698b64 \
     --hash=sha256:01582723b3ccd6939ab7b3a78622c573799d5d8737b534b86d0e06ac18dbde4a \
diff --git a/deps/pip/requirements_3_12.in b/deps/pip/requirements_3_12.in
index 1bb6e8b9..a39491fc 100644
--- a/deps/pip/requirements_3_12.in
+++ b/deps/pip/requirements_3_12.in
@@ -17,7 +17,7 @@
 -r requirements_tests.in
 -r requirements_ncore.in
 
-# Public API restrictions for 3.12
+# Public API restrictions for 3.12: most recent zarr-python 2
 zarr>=2.18.7,<3.0.0 # zarr-python 2.18.7 is the first version with the correct numcodecs<0.16 restriction
 
 # torch: same version as the python 3.11 environment
diff --git a/deps/pip/requirements_3_13.in b/deps/pip/requirements_3_13.in
index 5c2627ea..14d3debc 100644
--- a/deps/pip/requirements_3_13.in
+++ b/deps/pip/requirements_3_13.in
@@ -17,8 +17,8 @@
 -r requirements_tests.in
 -r requirements_ncore.in
 
-# Public API restrictions for 3.13
-zarr>=2.18.7,<3.0.0 # zarr-python 2.18.7 is the first version with the correct numcodecs<0.16 restriction
+# Public API restrictions for 3.13: most recent zarr-python 3
+zarr>=3.4.0 # most recent zarr-python 3 (python>=3.12)
 
 # torch: most recent version
 # CPU-only on purpose: python 3.12 / 3.13 cover the CPU code paths, CUDA is tested with python 3.11
diff --git a/deps/pip/requirements_3_13.txt b/deps/pip/requirements_3_13.txt
index 131dc3cd..29ff157c 100644
--- a/deps/pip/requirements_3_13.txt
+++ b/deps/pip/requirements_3_13.txt
@@ -3,9 +3,6 @@
 --index-url https://pypi.org/simple
 --find-links https://download.pytorch.org/whl/cpu/torch
 
-asciitree==0.3.3 \
-    --hash=sha256:4aa4b9b649f85e3fcb343363d97564aa1fb62e249677f2e18a96765145cc0f6e
-    # via zarr
 cbor2==6.1.5 \
     --hash=sha256:015ed73f10e1f7b67306d41e36e0d7dc40e4a2100bc5c29b7a7f039ad3dc9061 \
     --hash=sha256:040cf628af473fe18cb6f56bdac556d2398102e56852aab5206fbeb3dbde6b52 \
@@ -77,13 +74,9 @@ dataclasses-json==0.6.7 \
     --hash=sha256:0dbf33f26c8d5305befd61b39d2b3414e8a407bedc2834dea9b8d642666fb40a \
     --hash=sha256:b6b3e528266ea45b9535223bc53ca645f5208833c29229e847b3f26a1cc55fc0
     # via -r deps/pip/requirements_ncore.in
-deprecated==3.0.0 \
-    --hash=sha256:16850204d3a1e6bb0acd06bff48d96e8b0a0d25d1c52f71705405a0f4894192d \
-    --hash=sha256:58204cf4a7f6270d547af5c278ee7a6bb56045a4b3d8441a1cd11660f41b7939
-    # via numcodecs
-fasteners==0.20 \
-    --hash=sha256:55dce8792a41b56f727ba6e123fcaee77fd87e638a6863cec00007bfea84c8d8 \
-    --hash=sha256:9422c40d1e350e4259f509fb2e608d6bc43c0136f79a00db1b49046029d0b3b7
+donfig==0.8.1.post1 \
+    --hash=sha256:2a3175ce74a06109ff9307d90a230f81215cbac9a751f4d1c6194644b8204f9d \
+    --hash=sha256:3bef3413a4c1c601b585e8d297256d0c1470ea012afa6e8461dc28bfb7c23f52
     # via zarr
 filelock==4.1.0 \
     --hash=sha256:2ce9818e3e2d8f284c1a964414447ef148d42a5fd5e2a477a7118e574b293ec1 \
@@ -95,6 +88,39 @@ fsspec==2026.9.0 \
     # via
     #   torch
     #   universal-pathlib
+google-crc32c==1.9.0 \
+    --hash=sha256:0568b17ed90ac596f29400d99e243fd0cc6276766183def888d1bf8d1dc13827 \
+    --hash=sha256:0abe7e202c25909869c35672ab0f2fe748a7acf276eb78577332a7c38999740f \
+    --hash=sha256:13179f7e3282617923e957b8e54b8f9c3968030f48640a9f47fd7c5c38c4a215 \
+    --hash=sha256:16865b477d7941712cb0e0aad8ad4815e984fb5fc16d3fdaef7d986e26e53c95 \
+    --hash=sha256:265233aff33d835f5b909584fe36ab29647b598c271b661a300001099109e53e \
+    --hash=sha256:280f3a3e47af0eeba3a3e5aa7d311af77001812b8df80fb8beafcd0b40eaf7f1 \
+    --hash=sha256:3abb18297d9ef0ab120531838be0e6d68c9fa876570e11c229c48f2edac23ce7 \
+    --hash=sha256:43a2dc26f9be213fbe0b4fc4a1088c5d45cbfcb3247420ccc820f0fc3edeea86 \
+    --hash=sha256:4488f1553a9ab7e86cdedc833374a7e904031803b995dc0bd0be48c271fa6556 \
+    --hash=sha256:457d0d9a4718fd52b1494eac5c200ad25beeadbdc91843d550a003910838589f \
+    --hash=sha256:51cb4e23a38ad4f495f35f87c233ca3ea6b9c4559e7ac383cdef786fab0f7977 \
+    --hash=sha256:53fdafef58e230d0c946ab5f8446d123d9f548230a73b29c8b41c9546f268bc1 \
+    --hash=sha256:56610f548f1b35c9568b9d1de30423480f505dae4991556072d5802820ff35c4 \
+    --hash=sha256:5695c8b9327e040b2aba12c6659b0acb5995314ef0af0192da66e662e011103b \
+    --hash=sha256:6a3b2c8a343c570ed8100a7627c20badfd92c6caa2067093a86be45af27f5b1b \
+    --hash=sha256:7b8c84c3d159ab6817fe3f74e6e6cef099c3f95dcec3abc0d8afb1404642efbe \
+    --hash=sha256:8535e75dfead304f30e9122b9ea2c0a570dbaa52c176a0a591540c7914c1e46d \
+    --hash=sha256:8583ec21d56b565d68ab2963cc7e21b3b271247c29b04286068255ef65f221bd \
+    --hash=sha256:86764b99e7a607830d93cb5b75e0ec3ff6cb06d3c274624418473cee701900d4 \
+    --hash=sha256:8b91f41645b15a720357183fa5716682ada441873e3c462c15f9714be36f146b \
+    --hash=sha256:af73200fa9791ccd380f3598235dba8d82b8af0905df045b3dc60b59836e8ddd \
+    --hash=sha256:ccfe40021fd6afe23361175cf7551e3cef5fd34dc1ebe319f14993a83579e0eb \
+    --hash=sha256:dee799544cae42a42b17a88e38b59cf2c271051dc001da2117a8ff240ffa0548 \
+    --hash=sha256:e6b529a6a287104ec79d281c411685231200ce954a29c28ab8e5093cb6e130fb \
+    --hash=sha256:e6e8be8a94436079cb5340f6d495d9d7ba30124d8b952703994c739c7c06e236 \
+    --hash=sha256:f1dc17d987ddcc5eba12a7ce48f0eb93141dea236b170c1101151396edf2f0cf \
+    --hash=sha256:f2b64641bca27497b986b9d87883014035aa904cb4fa333407c6752b3afee9ba \
+    --hash=sha256:f894a2877650b56201d26a012a257b76d54a68834dc3913a93830ca8a047b075 \
+    --hash=sha256:f97c3806dcea41c29c04965347b0e12481561b75e0045dc7a4f69d75dec5d9b1 \
+    --hash=sha256:fb63a8d7fa2e95dcff1ca16af2f4d88b526fa5ff72d1696285884ac2d49b6963 \
+    --hash=sha256:fbef61a3794e011c65fb4396a196cf123a7f474fe5a443db8e5dd7d751b9e6d4
+    # via zarr
 iniconfig==2.3.1 \
     --hash=sha256:67f4b9c50da0dedf52af349e7749a80a9057a5031199791b906c3bb3ae878960 \
     --hash=sha256:9121e2c1fdb355232495be3194c8dfe87ccc2d5dee45947b78e68f499790d7a7
@@ -264,6 +290,90 @@ mpmath==1.3.0 \
     --hash=sha256:7a28eb2a9774d00c7bc92411c19a89209d5da7c4c9a9e227be8330a23a25b91f \
     --hash=sha256:a0b2b9fe80bbcd81a6647ff13108738cfb482d481d826cc0e02f5b35e5c88d2c
     # via sympy
+msgspec==0.22.0 \
+    --hash=sha256:0067057df265795f742658b15dbe53f3b6f21d19dcfa53676db11088cfa41e0a \
+    --hash=sha256:024138c51afd335d0b4dce401be33902caafac2b64f8c9f2509a378986175d98 \
+    --hash=sha256:05dbc8268e50c9232ec72b9af1c7b13049aade4d1197764e38c427048706e046 \
+    --hash=sha256:0666a1520cab86796612e794e71107e0fbf5e8ff3ddcdfcfff8f1d94b860d2f1 \
+    --hash=sha256:0739b068f31f2004a364f97679ba91f2f5ecd6ec2a5b4b890188ab5c57d20672 \
+    --hash=sha256:08826f5e5b0fa2f7a88592c396a243cfcc63d37e19f9d4fbe3b3f1be2fbdc404 \
+    --hash=sha256:0922714feff5300aacd8ecd65fa828317ce4bf5212b3139258c0bfc0253cd80e \
+    --hash=sha256:0a13624a4969159fe35d8c2a3d377b2b61bbd8585e327440d5e52725affcce38 \
+    --hash=sha256:0b25dcbc108783cb72503ed705b9fbb8c3cb02ee5801923f44b5f038c91cc365 \
+    --hash=sha256:0b31746da07cba0e330c6433a94a4699ad77d3aeb9638d1a320a7686b69f6249 \
+    --hash=sha256:0dfadea8bdcfafc614bd031de55a8ede22b43445cfff6d8b77cc0c07d3edc8a8 \
+    --hash=sha256:10d0d1d464960d99a949f7ca01ef8928e51c472433a5f5ab74b2d695fb830652 \
+    --hash=sha256:12a887c4c06e4a771a2db32c9a80c7bb21866b12458025f636dcdc2253331c28 \
+    --hash=sha256:1e547966017265c0d23342bcf2e027305dde40ea042d16694a9b96b4f696a052 \
+    --hash=sha256:21460f54cee9208239b1a8421fdf25bffc77293e1daba88f585711ad839b9758 \
+    --hash=sha256:21c887d4de397355f6635c2a037b1c067882dac5d132a1793d63bbf7cf5ca78e \
+    --hash=sha256:221cbcbfa4478152b91d37dcfd4830e2be92773e8139e883f43773450ebacef8 \
+    --hash=sha256:263e110955ed76fe0af2d79f819903b50a70dc0e7a752eb7aabe79d2e0a084fb \
+    --hash=sha256:268594d0bae5510572599a6ab0364dd9de43c867d24a30856cd9f5edb63d8dc6 \
+    --hash=sha256:27d9ef46c80884f9c4f323e0b18bec464287e872121e70f2cbe47335780bf597 \
+    --hash=sha256:28f53f3604dd3e70225f7563c831628dbb03299b428f8e62aadb4b628e386874 \
+    --hash=sha256:38c5b9bd347bc9abbcee40752be3c5117854e891ea7a1881a56d4b3dec58c5e7 \
+    --hash=sha256:38f7022fbe91954b31afe3888a0af1b652e0f370fafdeb1d425f4a814d789c9f \
+    --hash=sha256:3c789b5ccd07c0a3c09767108ee06e089b2875f2309a4569c2648f30a8d31dfa \
+    --hash=sha256:3ca7d4cd69fbb66bd2da6211d3e79d40542d196c16c6d99bf838f76767ad35be \
+    --hash=sha256:4600dbec738ed74e4c9bd35503e84701200ea7db344cfdeda80677b3ee53eb64 \
+    --hash=sha256:4a663a8d7f6ad56ac1dbcba91e046ba8ebab7773ae72ef3dd3c47f8226919184 \
+    --hash=sha256:508278300dd4efbd21cd3a4b2b016160a5feac98bc880d3673f6c06697baaf62 \
+    --hash=sha256:57c282f474e17acf6bcf84f393c73afd45d6eba47cccff8b76b79c4fbb8a3b54 \
+    --hash=sha256:5aa24eb475d070ecbbe5b21080fc3ce4b0b76c60de25cfe0c9678d8fb44bb42f \
+    --hash=sha256:5e4f7e09cceac7dbf4c0761b8ae7df51c55b5df5e9af7aff2c895aac1ebea015 \
+    --hash=sha256:614e2c827e0a3f934f3cf0cf4ba65210df8132b75a69a8a1f51bb3b2caf0ac5a \
+    --hash=sha256:627bfdfe5a4b3d916b3360b30f4cddeee3a084f56593e33527c6872fa8322ff9 \
+    --hash=sha256:65eea14bc65ccfeb8f3af62cb204841871e2961f002d7fa87dbe0f79dacf1c1c \
+    --hash=sha256:6ad64f5c260866b0d543f89f50cee43628989c1433c5de7ce820281fa28a2611 \
+    --hash=sha256:6ae370f92f3517f0e6f209ba7cc649c957b444868439197e046be07154667551 \
+    --hash=sha256:6f48317f05312bfdf78248f53933f830f07ab75cc1c813ac3ca4220cb3b5b019 \
+    --hash=sha256:71cbbdb39631064e2f2f9e9ac2b1b69931d72276eb5f9da4ed025726296bdbb6 \
+    --hash=sha256:7293dee54de040cfa225c22151cc3d72f17cd674b5ebcb52f38fb9f5701592e6 \
+    --hash=sha256:749899563d26b211379f142b8ffd7e2d7da149a51717798f0ce994dce50324f0 \
+    --hash=sha256:7c1e76c6bd523141b9c05c2f8a70979cd0efedbd68855a66f292f8892c0b8fc7 \
+    --hash=sha256:884c28c80b0a511595b29a9b04a3a230c3797369e4a033e6d5c6d9b5427f8e09 \
+    --hash=sha256:885c6e0c89d6103648525fe62aa78d600054dedf7b3713d23b15d7ddb6d66a13 \
+    --hash=sha256:8c8e84789918fbc15a503b92a829115ddd7567ecd3e4778bd418c56abbb86c11 \
+    --hash=sha256:8d67582478b0eaabb899f2fb255c878ee7de57dff80eb73ab24f1865524ec441 \
+    --hash=sha256:8f0a5c25516e2034b2db7767081759ff8996e214def9c43b3055f61e1be1caad \
+    --hash=sha256:99c401861c5bb3a57f7d6423ea7ed4352cd57aa3f04f4fbe9f3e3e4564a10f08 \
+    --hash=sha256:9a696f23f7c1ffb31fae308502e01a3965c3891d5c400f01d0d1096dbe77519e \
+    --hash=sha256:a1dab6a99c759d1391ab2993388c1892746a697254f4b5dc6c059ca6e3bfbc8b \
+    --hash=sha256:a52eba5c9528fd181fcec39d22b67aaa1dccc6cfe8e24d3f5d41130e6d04289d \
+    --hash=sha256:a66b1766311e42371e509c996c3933b161c7ae0eabdf361af5316dec197e1022 \
+    --hash=sha256:a6c8a3f210421e29d8f7e9815f106cf59d758665b7fe5428e61152ce24fe65d7 \
+    --hash=sha256:a6db3806b3b76ca78064255eac6fa101a8a64fe6f698d80fbaf81fdfa21217d4 \
+    --hash=sha256:a88d939d3fe4b8c7314645ebcd6e86c8c8a512ea7820d6550355973e803bc0f1 \
+    --hash=sha256:a8b98ae215a102cbf6635f7df45f5c4af12f77fad1f7b71b9808fcf868a5735d \
+    --hash=sha256:ab1e9e7531e353653b906cdd12a0220cc288a1e8e3436aabc65f4508d91b14d9 \
+    --hash=sha256:b3113ebcceeb7693a915183c73d92c10bf5c62851dd187cab43bd025fb587419 \
+    --hash=sha256:b5a169b5b03f0f2c7a296c002647db1dab75d2cd501bca34e32b71cab0261b56 \
+    --hash=sha256:b60b43425a47eb9cfe987f6874e354ca7c760e58e295b4e2273ff03574df28a1 \
+    --hash=sha256:b6d3ca19a8ff28d0a67a1824e2bff7ec649ec795c80a265f20ade4caa63080de \
+    --hash=sha256:b962000e11dd34fb210a5a2c57a8a62b2d92b381c8cb3b05c075a83e38f8d645 \
+    --hash=sha256:bc374dedd5f85a5f4de2386dc5f737894ccb8c1ac18e9566ce66fd9839e6285d \
+    --hash=sha256:c3c510aba9015c085e514b75a9b3f1ed7c4591ae5e379655821b8bba51f30cc7 \
+    --hash=sha256:c6c310ef83e7e291b01a63298828f848348bb99e84a1098c4b3923c05674d032 \
+    --hash=sha256:c6f06576eced70462179a4b4638e84cf69fdbba37f44d13a64a21739c131a830 \
+    --hash=sha256:cfc3d9557de9c806318725b702f3e664db33167bb42892079b693c69893fd33b \
+    --hash=sha256:d2f950239ff1fc7322c6f9634807310265149cb168270d3ddcdda5b6ada13a28 \
+    --hash=sha256:d7a738826936c72348c613061d260446f13c82b6fd7d5d7705b6911ab8dca2f3 \
+    --hash=sha256:dce29a04966e31abf9b83b697c6d672486526dc5d03fcd6970cb56d5dc1fbeea \
+    --hash=sha256:dd9568695911055440d2bb7099ed9098fc181d335daa772d0eb3fe8f31ba4efb \
+    --hash=sha256:e0aa0cc3f18c35bab79bd7b87fde95d6274a9deddeebd1ea541f8066a5073165 \
+    --hash=sha256:e79725246291516a7359caad5fb743ddc0ec66ed40d2381fb846325b5031504e \
+    --hash=sha256:ebd211d7af79ed8710c64e9e8d4c0d02749bc20170e7ab4e1c5801ca7c99d25b \
+    --hash=sha256:ec108e96fdaa8fdbe5bb993ec97a9d1faa69b3a521eecd71a6e5acbe0e29ae69 \
+    --hash=sha256:f039ef5207b847f075a0a43020ee6140cd47505f890e47e157f2deb485c2dc96 \
+    --hash=sha256:f13c127a945479bc9db057eb253b8851075c8e1ae07ffc967bfa1c5676203a86 \
+    --hash=sha256:f2ddea9d78d09460f06c26a7a508adcd049761c3208776162b8eb79b8a032cff \
+    --hash=sha256:f3413e3647275f787b21b4dfb4836a59a1a5acf1018ab1d45843b1d7edf15c22 \
+    --hash=sha256:f7a923bcde480065c8e25967464cfb2a687ee67000bb43157e2d57e40eca7305 \
+    --hash=sha256:fa3689b9dfcc663358ef23ba4299d7460f01108515b041a7d30d05908ac9c32f \
+    --hash=sha256:fb1e129b81ac8fcf9ec649b081c6c8da1c7ea6f87cab336d46386abc2cd855c1 \
+    --hash=sha256:feafe612034d49e9144340c0b5168ee4e22c2af4aaa2c1db11ae84e1aac9543b
+    # via zarr
 mypy-extensions==1.1.0 \
     --hash=sha256:1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505 \
     --hash=sha256:52e68efc3284861e772bbcd66823fde5ae21fd2fdb51c62a211403730b916558
@@ -272,20 +382,28 @@ networkx==3.7 \
     --hash=sha256:e3fd2c13a7814cee3746340d8d7f8598a67f16a58bf47fb7f8793fab6efca1b0 \
     --hash=sha256:fd77a511bd90f39f3d016351345b52cf5319b813bdca01de3f755d3cca62e96a
     # via torch
-numcodecs==0.15.1 \
-    --hash=sha256:1d471a1829ce52d3f365053a2bd1379e32e369517557c4027ddf5ac0d99c591e \
-    --hash=sha256:1dfdea4a67108205edfce99c1cb6cd621343bc7abb7e16a041c966776920e7de \
-    --hash=sha256:698f1d59511488b8fe215fadc1e679a4c70d894de2cca6d8bf2ab770eed34dfd \
-    --hash=sha256:a34f0fe5e5f3b837bbedbeb98794a6d4a12eeeef8d4697b523905837900b5e1c \
-    --hash=sha256:a4f7bdb26f1b34423cb56d48e75821223be38040907c9b5954eeb7463e7eb03c \
-    --hash=sha256:b0a9d9cd29a0088220682dda4a9898321f7813ff7802be2bbb545f6e3d2f10ff \
-    --hash=sha256:bef8c8e64fab76677324a07672b10c31861775d03fc63ed5012ca384144e4bb9 \
-    --hash=sha256:c3a09e22140f2c691f7df26303ff8fa2dadcf26d7d0828398c0bc09b69e5efa3 \
-    --hash=sha256:cdfaef9f5f2ed8f65858db801f1953f1007c9613ee490a1c56233cd78b505ed5 \
-    --hash=sha256:daed6066ffcf40082da847d318b5ab6123d69ceb433ba603cb87c323a541a8bc \
-    --hash=sha256:e2547fa3a7ffc9399cfd2936aecb620a3db285f2630c86c8a678e477741a4b3c \
-    --hash=sha256:e3d82b70500cf61e8d115faa0d0a76be6ecdc24a16477ee3279d711699ad85f3 \
-    --hash=sha256:eeed77e4d6636641a2cc605fbc6078c7a8f2cc40f3dfa2b3f61e52e6091b04ff
+numcodecs==0.17.0 \
+    --hash=sha256:0fc125d1c726c1937cde346e109e3662a2b4ff6be073289da7d124d172aceda5 \
+    --hash=sha256:1837d4d1d646cecd3ab2d1ba22956295d709edea0bddc952737c647bec1d03c4 \
+    --hash=sha256:1cce4bf2278ed74841c2088acfd38e67c3e5aa77e3bc1962ef0fa2931becbb12 \
+    --hash=sha256:1ebd63cdb8985c66257bc037fcdff5f38637aff72d7ef62612ec46f2299e8749 \
+    --hash=sha256:2e29732c5e3a83663e51b40007819d8fd0aae16a2322f7044ce13a2460a99e23 \
+    --hash=sha256:3e4e351566b3ab2f6255a9d91c6c48e1d0f9ec6e2ae409a148e091a8fc0a80b0 \
+    --hash=sha256:4a62e5a821ccfbe425bbdd9a079f8b6c41b7e796ff3c99324530561193a53047 \
+    --hash=sha256:4c36f6fd14dc22939172145c24d3b3eab2410c34ed807906a5ece5f4541c7c43 \
+    --hash=sha256:4d83befe67a51ba6a988c562209bf13836438c1b6dce23049d84ff42854af32d \
+    --hash=sha256:4f43ba0d834ce012ed482996a7424df9077a47d5899ede2d1d54fe85e6eb12fa \
+    --hash=sha256:657b1f9aa4b1025aa0fa7d4bd8d7492900950a11f636dff622bd208c0b99e35e \
+    --hash=sha256:69b9b4685097c4d478a0c829debf4470555ec63e92cdd2c6b5f195460f1dc888 \
+    --hash=sha256:6c3342d91ed7cf59c1be84396edd364e936bb0ec9e366d24bb69689748d19625 \
+    --hash=sha256:6f1293581326e92293b142bd05b389f6682ed1ce333f36f116344bca340cfd10 \
+    --hash=sha256:7065b3349b73d54785aa89e00d0b97d80f664e9056757929d28151f9208dc04c \
+    --hash=sha256:8697a4631fedded77a75d333e4926b1eb3a11bc7d3e30213e7e565d6910526d0 \
+    --hash=sha256:a854e9c89f58eeeb2453f3c1637d1916797edb6eaff26bc186a6cdb09d187092 \
+    --hash=sha256:d30c69b4bdb1755af1022fa913e184eaadc4fc0cd38f736e483e8ad205e130d1 \
+    --hash=sha256:de2c66db238e74e66fe9be7e02b7e0129b75d3f812d38e4019eb0102cc2dcdf0 \
+    --hash=sha256:e8db2e337bdafd3bb5f891a2543b53b2b36a509ce9d587af2846db3715b6c8b9 \
+    --hash=sha256:ecd0f6a10e3f8afbbb16ecc999d2b06aa2a31a2946f1c1a85d15d91a1ebcfef3
     # via zarr
 numpy==2.5.4 \
     --hash=sha256:067374eb538c34c745436365cf7b0112595c1d326f21ce4ff340f61230239fbb \
@@ -377,6 +495,7 @@ packaging==26.3 \
     # via
     #   marshmallow
     #   pytest
+    #   zarr
 parameterized==0.9.0 \
     --hash=sha256:4e0758e3d41bea3bbd05ec14fc2c24736723f243b28d702081aef438c9372b1b \
     --hash=sha256:7fc905272cefa4f364c1a3429cbbe9c0f98b793988efb5bf90aac80f08db09b1
@@ -486,6 +605,81 @@ pytest==9.1.1 \
     --hash=sha256:1088fbde8f2b49d95a549a195707afa7a76a3ce9bcadc26b6d71f0ffda5fe313 \
     --hash=sha256:37a86b45efb9a47a61a36449063e8e18d0cab3161329fc099eb21783169c4f0c
     # via -r deps/pip/requirements_tests.in
+pyyaml==6.0.3 \
+    --hash=sha256:00c4bdeba853cc34e7dd471f16b4114f4162dc03e6b7afcc2128711f0eca823c \
+    --hash=sha256:0150219816b6a1fa26fb4699fb7daa9caf09eb1999f3b70fb6e786805e80375a \
+    --hash=sha256:02893d100e99e03eda1c8fd5c441d8c60103fd175728e23e431db1b589cf5ab3 \
+    --hash=sha256:02ea2dfa234451bbb8772601d7b8e426c2bfa197136796224e50e35a78777956 \
+    --hash=sha256:0f29edc409a6392443abf94b9cf89ce99889a1dd5376d94316ae5145dfedd5d6 \
+    --hash=sha256:10892704fc220243f5305762e276552a0395f7beb4dbf9b14ec8fd43b57f126c \
+    --hash=sha256:16249ee61e95f858e83976573de0f5b2893b3677ba71c9dd36b9cf8be9ac6d65 \
+    --hash=sha256:1d37d57ad971609cf3c53ba6a7e365e40660e3be0e5175fa9f2365a379d6095a \
+    --hash=sha256:1ebe39cb5fc479422b83de611d14e2c0d3bb2a18bbcb01f229ab3cfbd8fee7a0 \
+    --hash=sha256:214ed4befebe12df36bcc8bc2b64b396ca31be9304b8f59e25c11cf94a4c033b \
+    --hash=sha256:2283a07e2c21a2aa78d9c4442724ec1eb15f5e42a723b99cb3d822d48f5f7ad1 \
+    --hash=sha256:22ba7cfcad58ef3ecddc7ed1db3409af68d023b7f940da23c6c2a1890976eda6 \
+    --hash=sha256:27c0abcb4a5dac13684a37f76e701e054692a9b2d3064b70f5e4eb54810553d7 \
+    --hash=sha256:28c8d926f98f432f88adc23edf2e6d4921ac26fb084b028c733d01868d19007e \
+    --hash=sha256:2e71d11abed7344e42a8849600193d15b6def118602c4c176f748e4583246007 \
+    --hash=sha256:34d5fcd24b8445fadc33f9cf348c1047101756fd760b4dacb5c3e99755703310 \
+    --hash=sha256:37503bfbfc9d2c40b344d06b2199cf0e96e97957ab1c1b546fd4f87e53e5d3e4 \
+    --hash=sha256:3c5677e12444c15717b902a5798264fa7909e41153cdf9ef7ad571b704a63dd9 \
+    --hash=sha256:3ff07ec89bae51176c0549bc4c63aa6202991da2d9a6129d7aef7f1407d3f295 \
+    --hash=sha256:41715c910c881bc081f1e8872880d3c650acf13dfa8214bad49ed4cede7c34ea \
+    --hash=sha256:418cf3f2111bc80e0933b2cd8cd04f286338bb88bdc7bc8e6dd775ebde60b5e0 \
+    --hash=sha256:44edc647873928551a01e7a563d7452ccdebee747728c1080d881d68af7b997e \
+    --hash=sha256:4a2e8cebe2ff6ab7d1050ecd59c25d4c8bd7e6f400f5f82b96557ac0abafd0ac \
+    --hash=sha256:4ad1906908f2f5ae4e5a8ddfce73c320c2a1429ec52eafd27138b7f1cbe341c9 \
+    --hash=sha256:501a031947e3a9025ed4405a168e6ef5ae3126c59f90ce0cd6f2bfc477be31b7 \
+    --hash=sha256:5190d403f121660ce8d1d2c1bb2ef1bd05b5f68533fc5c2ea899bd15f4399b35 \
+    --hash=sha256:5498cd1645aa724a7c71c8f378eb29ebe23da2fc0d7a08071d89469bf1d2defb \
+    --hash=sha256:5cf4e27da7e3fbed4d6c3d8e797387aaad68102272f8f9752883bc32d61cb87b \
+    --hash=sha256:5e0b74767e5f8c593e8c9b5912019159ed0533c70051e9cce3e8b6aa699fcd69 \
+    --hash=sha256:5ed875a24292240029e4483f9d4a4b8a1ae08843b9c54f43fcc11e404532a8a5 \
+    --hash=sha256:5fcd34e47f6e0b794d17de1b4ff496c00986e1c83f7ab2fb8fcfe9616ff7477b \
+    --hash=sha256:5fdec68f91a0c6739b380c83b951e2c72ac0197ace422360e6d5a959d8d97b2c \
+    --hash=sha256:6344df0d5755a2c9a276d4473ae6b90647e216ab4757f8426893b5dd2ac3f369 \
+    --hash=sha256:64386e5e707d03a7e172c0701abfb7e10f0fb753ee1d773128192742712a98fd \
+    --hash=sha256:652cb6edd41e718550aad172851962662ff2681490a8a711af6a4d288dd96824 \
+    --hash=sha256:66291b10affd76d76f54fad28e22e51719ef9ba22b29e1d7d03d6777a9174198 \
+    --hash=sha256:66e1674c3ef6f541c35191caae2d429b967b99e02040f5ba928632d9a7f0f065 \
+    --hash=sha256:6adc77889b628398debc7b65c073bcb99c4a0237b248cacaf3fe8a557563ef6c \
+    --hash=sha256:79005a0d97d5ddabfeeea4cf676af11e647e41d81c9a7722a193022accdb6b7c \
+    --hash=sha256:7c6610def4f163542a622a73fb39f534f8c101d690126992300bf3207eab9764 \
+    --hash=sha256:7f047e29dcae44602496db43be01ad42fc6f1cc0d8cd6c83d342306c32270196 \
+    --hash=sha256:8098f252adfa6c80ab48096053f512f2321f0b998f98150cea9bd23d83e1467b \
+    --hash=sha256:850774a7879607d3a6f50d36d04f00ee69e7fc816450e5f7e58d7f17f1ae5c00 \
+    --hash=sha256:8d1fab6bb153a416f9aeb4b8763bc0f22a5586065f86f7664fc23339fc1c1fac \
+    --hash=sha256:8da9669d359f02c0b91ccc01cac4a67f16afec0dac22c2ad09f46bee0697eba8 \
+    --hash=sha256:8dc52c23056b9ddd46818a57b78404882310fb473d63f17b07d5c40421e47f8e \
+    --hash=sha256:9149cad251584d5fb4981be1ecde53a1ca46c891a79788c0df828d2f166bda28 \
+    --hash=sha256:93dda82c9c22deb0a405ea4dc5f2d0cda384168e466364dec6255b293923b2f3 \
+    --hash=sha256:96b533f0e99f6579b3d4d4995707cf36df9100d67e0c8303a0c55b27b5f99bc5 \
+    --hash=sha256:9c57bb8c96f6d1808c030b1687b9b5fb476abaa47f0db9c0101f5e9f394e97f4 \
+    --hash=sha256:9c7708761fccb9397fe64bbc0395abcae8c4bf7b0eac081e12b809bf47700d0b \
+    --hash=sha256:9f3bfb4965eb874431221a3ff3fdcddc7e74e3b07799e0e84ca4a0f867d449bf \
+    --hash=sha256:a33284e20b78bd4a18c8c2282d549d10bc8408a2a7ff57653c0cf0b9be0afce5 \
+    --hash=sha256:a80cb027f6b349846a3bf6d73b5e95e782175e52f22108cfa17876aaeff93702 \
+    --hash=sha256:b30236e45cf30d2b8e7b3e85881719e98507abed1011bf463a8fa23e9c3e98a8 \
+    --hash=sha256:b3bc83488de33889877a0f2543ade9f70c67d66d9ebb4ac959502e12de895788 \
+    --hash=sha256:b865addae83924361678b652338317d1bd7e79b1f4596f96b96c77a5a34b34da \
+    --hash=sha256:b8bb0864c5a28024fac8a632c443c87c5aa6f215c0b126c449ae1a150412f31d \
+    --hash=sha256:ba1cc08a7ccde2d2ec775841541641e4548226580ab850948cbfda66a1befcdc \
+    --hash=sha256:bdb2c67c6c1390b63c6ff89f210c8fd09d9a1217a465701eac7316313c915e4c \
+    --hash=sha256:c1ff362665ae507275af2853520967820d9124984e0f7466736aea23d8611fba \
+    --hash=sha256:c2514fceb77bc5e7a2f7adfaa1feb2fb311607c9cb518dbc378688ec73d8292f \
+    --hash=sha256:c3355370a2c156cffb25e876646f149d5d68f5e0a3ce86a5084dd0b64a994917 \
+    --hash=sha256:c458b6d084f9b935061bc36216e8a69a7e293a2f1e68bf956dcd9e6cbcd143f5 \
+    --hash=sha256:d0eae10f8159e8fdad514efdc92d74fd8d682c933a6dd088030f3834bc8e6b26 \
+    --hash=sha256:d76623373421df22fb4cf8817020cbb7ef15c725b9d5e45f17e189bfc384190f \
+    --hash=sha256:ebc55a14a21cb14062aa4162f906cd962b28e2e9ea38f9b4391244cd8de4ae0b \
+    --hash=sha256:eda16858a3cab07b80edaf74336ece1f986ba330fdb8ee0d6c0d68fe82bc96be \
+    --hash=sha256:ee2922902c45ae8ccada2c5b501ab86c36525b883eff4255313a253a3160861c \
+    --hash=sha256:efd7b85f94a6f21e4932043973a7ba2613b059c4a000551892ac9f1d11f5baf3 \
+    --hash=sha256:f7057c9a337546edc7973c0d3ba84ddcdf0daa14533c2065749c9075001090e6 \
+    --hash=sha256:fa160448684b4e94d80416c0fa4aac48967a969efe22931448d853ada8baf926 \
+    --hash=sha256:fc09d0aa354569bc501d4e787133afc08552722d3ab34836a80547331bb5d4a0
+    # via donfig
 scipy==1.18.1 \
     --hash=sha256:011413b7426b75012840e35649e00fe0a2c3bae89fed433876e3a99251572efc \
     --hash=sha256:0ac49ea97594532dd44b7136094d35f5440fa06e6d9c6384a74c01764df388c5 \
@@ -566,8 +760,10 @@ typing-extensions==4.16.0 \
     --hash=sha256:dc983d19a509c94dba722ee6abd33940f7c05a89e243c47e907eb4db6f1a43e5
     # via
     #   -r deps/pip/requirements_ncore.in
+    #   numcodecs
     #   torch
     #   typing-inspect
+    #   zarr
 typing-inspect==0.9.0 \
     --hash=sha256:9ee6fc59062311ef8547596ab6b955e1b8aa46242d854bfc78f4f6b0eff35f9f \
     --hash=sha256:b23fc42ff6f6ef6954e4852c1fb512cdd18dbea03134f91f856a95ccc9461f78
@@ -576,112 +772,9 @@ universal-pathlib==0.3.10 \
     --hash=sha256:4487cbc90730a48cfb64f811d99e14b6faed6d738420cd5f93f59f48e6930bfb \
     --hash=sha256:dfaf2fb35683d2eb1287a3ed7b215e4d6016aa6eaf339c607023d22f90821c66
     # via -r deps/pip/requirements_ncore.in
-wrapt==2.5.0 \
-    --hash=sha256:05246a100da68259af521b88788f131ba005465f1c95d358cc3c03ec5e351b52 \
-    --hash=sha256:05b7a271e21598694dcb7a6d6224f04e80ecc24b00a31ec57c658196377b0b29 \
-    --hash=sha256:0810e060e58f7960405172ad21080df8e7335841c9fe97417bd7d3f05af24f90 \
-    --hash=sha256:0a7a369e7fca9fc8c2c50df634382009b09af416853da3d4515e4bb048a5b9ee \
-    --hash=sha256:0cabb9c17ab79b2549d1f23b36f436473ad9253ef53995a817feba26fae69d5b \
-    --hash=sha256:0d245ac03f5ae77f1eea6eb19edd9e778c2f772490c20496c2f1cd3a102ee1b6 \
-    --hash=sha256:0dfc38cb672af51fc29696ba9c6f05d2315f5e62c4af2564e50f07f81198a163 \
-    --hash=sha256:0ea62bc142f4fa8b2e0ab058f50699ccd679ef6199c8fa3cc1c2396c7a659000 \
-    --hash=sha256:107eea1a511e98a3a5033b0c2cb403fbb37f05dee6ac1fb85c0460d311ec278c \
-    --hash=sha256:1122f4f9e363da804ccba05a9f0f39baa3716eb82452c929258bc3f1420c899b \
-    --hash=sha256:148052fc55013930217f531c6978e918ab210a12bd73cc9bd6de661a7adaf620 \
-    --hash=sha256:181a45000506a6382eb337354ca7e8525690702f1ccf2eae4f23a210ff339543 \
-    --hash=sha256:18baaf966bdc22dbdbe6e4323da8c17c0b7e8ea941dccf5426e741ae398953d9 \
-    --hash=sha256:1a3c4035d2026b87ef23dd8d165f1f8d3853ee2bd02791cfd22bd8c6226c41ce \
-    --hash=sha256:1b35ef7379323a149a6398f6261248bf48b61666210bd69e2ab24a9a9afdedc0 \
-    --hash=sha256:1ccd22ef8690ca302425f26d4c3e8b24f1284d4351d20e49ce21f9c0dd58d64d \
-    --hash=sha256:1d4da5f0e9a719471502b0db80d5c97503aeca796b7aeb9ab8f47403b2be76e6 \
-    --hash=sha256:1ebc0d09906057ada57a32158a657364ca40b8f86e604da3e7d979069b601502 \
-    --hash=sha256:1f17c5a3836397bf59fd57b0e5b0dc42969b1daa70d31aa361c6e13cbf138b5a \
-    --hash=sha256:1ffb2823c95dbeb8a47fedfba9636b2afeb0a8ef94b66df97bd081bdfe5a263f \
-    --hash=sha256:21cfe343ef9c2deb865ad0d5c57822266447c88dcf6d8805dd8c363fe367f30c \
-    --hash=sha256:23a9d6cb6413359b76f030d6bbb75340b7669c69da245dde2919a4c93708993b \
-    --hash=sha256:2ac82ef59ee05e259902bc7cf73dee5e6397845e8ccdc376d9d25536b59a877c \
-    --hash=sha256:2e3d110d7f99644946f249927c346d9dba507a78815d1bcebdbf0d94c14c5649 \
-    --hash=sha256:2fd8a61c31220840c7f52621cf51c961af5058bd009a55bcdf4a6732bdb13b35 \
-    --hash=sha256:30cf86d1e35a57e772e4709a137c1731751255439368a41b1a2519561e385f4e \
-    --hash=sha256:36703cafc2ec059e118c2175e6cb7ad7299c2924aecdcb1b7a7ebbf7a3e20c19 \
-    --hash=sha256:3f5dfb867d3f58fde0d850a302f1164bb83f4a922b037882bcc7e0474aa8312f \
-    --hash=sha256:425349a99b8c9540399d36620c376dc26e6aca93071cd6fafa239c2f1b5d53a4 \
-    --hash=sha256:4343880acd72e74233baf092285aaaf4306244e31d7601828bd2600316027df0 \
-    --hash=sha256:439523506fd0d9af4f76d75f2e46aa4979e164fecba6890859326ff06edbaf9c \
-    --hash=sha256:47abb2bb7f15b416e72fbe5e68e49a6f09331dae6af1ca6055f5aa2251d2bd2f \
-    --hash=sha256:513dd1f4a1f91030d5656d9f4b3af8aa490d4db1eb53481e6846b8a5ee7acfcd \
-    --hash=sha256:5375ff1d2159e2ef847449e3dd2329441423d0a3d895d123273a67095fe7ca0c \
-    --hash=sha256:5421acb5c363a9bc959122a8645e3f1f42010c932dc53885b11a5ff5b5a6d730 \
-    --hash=sha256:557ebf4ce5568588368675014a2540405db687c2e4c7ad1eb83aa7e857be1864 \
-    --hash=sha256:57fa1a3fd1279b3ca7655b943ad61d298f2a2464a4cdca7ff298058e408322f9 \
-    --hash=sha256:58b2a87c65cbfb20917ec48ace47f71b1962c1f81dbf18a4052e3037abf72028 \
-    --hash=sha256:5df50c6133a63071fb77a0a091ceae36e78467b728d65e711ded02510edd9855 \
-    --hash=sha256:6058e12e9caa33468f9a36fb88c15a4bb30a479f997b37834b83abdbf062f264 \
-    --hash=sha256:6119e5d5bda268171af1f68ab5e7544c618f02a9bdc7c3929a458b719b88cd03 \
-    --hash=sha256:61253431e5a6f0eaeae71e2a101753fbaf4360b3e4dd8e118177d565ae119d40 \
-    --hash=sha256:6269637d9a54990430b4a769df15833935a46c4d004d9fe8a153bbadf0b9a097 \
-    --hash=sha256:63e58f96849f622ce769dcd705f83c7445cd9829bce1dae00e78bb031aec8096 \
-    --hash=sha256:65f2ee406dc592a5b22a7dc6abac13e8a3e8de4b2ecf5dc3c22937865496e4b6 \
-    --hash=sha256:672dd1bab4256311db1520b1b50e7a10cbaeaae2b0ac6bc5d858cc387ee605a2 \
-    --hash=sha256:6761765cc520ff9616fb035c02a85d1d744f7f70edd4649718fd0d09c589eacf \
-    --hash=sha256:691671ea05684f921ffc2e935fd3f9311c1795a10fbbfa006b46269733668f66 \
-    --hash=sha256:6a31cc61ad1be4091f5b094c91d53d206609928af13b7b24573cb857a3bc07fc \
-    --hash=sha256:6a9ee62a970075738909909bdbef3a7da9f7ae03dfca584db283547a29503b56 \
-    --hash=sha256:7138b0e7990e5555a905c519e8dad17c1da1f20e08b283e202c414229065740f \
-    --hash=sha256:734ea79e4707751fb7cc4f716123b2115cd16ed0c4926e803540aece22e3ba67 \
-    --hash=sha256:76fb341d5a707a4f211631b8c77259b2df149147e9d9c245ae6ba3dd936bfdfb \
-    --hash=sha256:77f0a74ff6f6cf89f5b673a732d5afe1911a6e6b1c017260836fdfdf85518dc1 \
-    --hash=sha256:78b7bdaa8b27b7f7607c66bdb6ab15c1dcbd9e9a1556a253a347dad511f615d1 \
-    --hash=sha256:79e68f0fd7d381b9bbd71776f602a2d5440d4d2077459128e02fd6607465422c \
-    --hash=sha256:7e25e9697f60af41fb86b08697e470b1e7eb6cd6ac0eb25e4b1f519839adc271 \
-    --hash=sha256:7e2af7ff9553c492f684a41d903ec41e37bf6267ee3206c17518a349201cae46 \
-    --hash=sha256:8837fbe708cb9d8a2d32a37dee836d24a531f02560db26418e2b181986fa21cb \
-    --hash=sha256:88bb24b9fdccb1d805258d5648533206eb58c58b6554985c47db53a89c11be85 \
-    --hash=sha256:94ebe745d1b0ebd3af91e22c38961616d32a228bdf9c326e7d4a6007a623d48e \
-    --hash=sha256:99f8ea48f14a71c5e2df8763e9a490e8af63096dcd67755b7bab0a4b74fc7cd7 \
-    --hash=sha256:a039b009693b58f7e0cf6121851218c2227223e7601cc24ed4957ba0990fd25b \
-    --hash=sha256:a10e9af5d4c5977d2d93e75d1662d076a8e8bf8c499dac7be4e78ebd8e0b59af \
-    --hash=sha256:a145a7826eddea3eb5814903f98f93756042b919bb5305544cb1331daa2705b1 \
-    --hash=sha256:a346408f19b6d589bf029f25f65c0b4cdeed6302ef8f40da4e5d1552d22dc037 \
-    --hash=sha256:a45a5249a6965d91aac9f991fda7c17e6b8b41fe91592a6099f182bf53c82724 \
-    --hash=sha256:a5bb346a34499e091e4fa23df58251ad192173c088e5413d893ca1c730c133c7 \
-    --hash=sha256:aae2f5f4c77335a39ebe5a1c77d5519f6cefefc7dbff50bd551e3771d927e3fc \
-    --hash=sha256:ab45839c912777e2738fed369636589c8b2a6d9c44ca56de0fd0814581d467c2 \
-    --hash=sha256:b312b3cc87951faaed3cfef984d768ee8bee7f935d9cc929aaa9946b0b96a98c \
-    --hash=sha256:b58138d19f34e32833e62de5e910bc2a8baae43310b921d783bd39b15227c2dd \
-    --hash=sha256:b620d7559b6b2197c5730332fab0867ecf1c8cb74d45533ebbcbcad1eacf4616 \
-    --hash=sha256:b640460f0ffb346b192686bd6fac5589e35a6c6640501c59a9fb6e82b0dd6bd8 \
-    --hash=sha256:b898caea081303006decc562c7fca5126f7c96507e78dd8f1ae3285dfa50ddc7 \
-    --hash=sha256:b95a6eca3b927853529eea958310563c83140ae8451dd5dc4399c7da385dc4f3 \
-    --hash=sha256:bad63bb4dea3c58e8078a3a173259ac2df5442a437e49c632b4099c0250e803b \
-    --hash=sha256:bc5607c1911c92530cb402ea90d818933cffb28bd8de9b453a2542279816d8c7 \
-    --hash=sha256:bc6491d3008ecabf685b0746f03ad8241a0950336939b14addb03af39b51a316 \
-    --hash=sha256:bcd42e7b69c8c1e33a29b79b28de03bdc08876a49745830f9162a3af860e06d0 \
-    --hash=sha256:c3a78a3161b3a9bf07725822fd24d379c1f3f161b6db49166c4131096ea73b4c \
-    --hash=sha256:c3dfb16e047c912e1a06bfbc45da752219f474897be85f55b8069f7936b08cc0 \
-    --hash=sha256:c48cdb6c904dca76d9915a579e4a5fab6b0c25f650c1019ce78a78effaf7a345 \
-    --hash=sha256:c57ddae24cf72eb6bd18112638a987cafe6109d90f2df111e6934362cc03ac1a \
-    --hash=sha256:c932273bc43b068538f3874fa5e6c2a60f33fa0b11c1ebc7768652f6a0608943 \
-    --hash=sha256:cbd45dfba6b5c1bfbabe1feb3c0f117fbd62416e98268d9a7cd9ad8802875356 \
-    --hash=sha256:ce4cab32c37ef71e69cf88f909b7febd0dd79543e5ae3650e2b874e0f3d3b975 \
-    --hash=sha256:d75d6366203c8d025c1a74bae0b565952187ddae79bd5c7bf10687652a56f020 \
-    --hash=sha256:def66258d97ebf1e4e97def12c5daa542d1cc728a3da83ed3a43933f56df6dab \
-    --hash=sha256:dfedf844892bf88f387cdeb9fa9bff1186904417c9f05087c99241bec24a65cc \
-    --hash=sha256:e0345d4c1f7aa5a27075d28a7f0e9ed386729198045f1f08b47f8320d6bbda23 \
-    --hash=sha256:e3c6fb1c1a516881353186bed9cfcb8899f968c03b3509720c79db0d967acf3b \
-    --hash=sha256:e5d4885c5625c9d2dcb49458700851574e9d0ea046c7a265526e990282f8ae8e \
-    --hash=sha256:e716f47c7f61e11709d3c0904213c94fc22999abf0c41461276cef886c1e8b4d \
-    --hash=sha256:ea4fdc79c0045d6bb1603c109127145245cafe888588888444e1e37fbeadbac3 \
-    --hash=sha256:f12e80c3089ebc03727d368f8205b811b5af2cd4a72b5e4cac75e901dd316e39 \
-    --hash=sha256:f42239c89430eee2d8a6dec39e34677abdbb67fff63caf2467dd6124ea4d4d58 \
-    --hash=sha256:f4ef4935962f7029b2058a99f1a47ccbffc3be919dddb3becb6c2c48eac3d9f0 \
-    --hash=sha256:fd91203e156d610ecb28b9ccd7b764af7a7b38662d7c163090babab0d10def0c \
-    --hash=sha256:fe09aac4837ec720af606a493e814dcc3f65631984e1b7231b589efcf9917024 \
-    --hash=sha256:ff909b934b1958e31784d412abab5cbb0709fdbc01c86f22965e1d15331371ba
-    # via deprecated
-zarr==2.18.7 \
-    --hash=sha256:ac3dc4033e9ae4e9d7b5e27c97ea3eaf1003cc0a07f010bd83d5134bf8c4b223 \
-    --hash=sha256:b2b8f66f14dac4af66b180d2338819981b981f70e196c9a66e6bfaa9e59572f5
+zarr==3.4.1 \
+    --hash=sha256:38b540578a119352bdce02a720d7bfd99846e77f0728c58b1bab3d1f547526a0 \
+    --hash=sha256:b34bda11ceb199c81ee78ecd42cd02f46c7f67a6d8a1e9bc501cafd5a1795356
     # via
     #   -r deps/pip/requirements_3_13.in
     #   -r deps/pip/requirements_ncore.in
diff --git a/deps/pip/requirements_ncore.in b/deps/pip/requirements_ncore.in
index 54bf97e8..d691be4c 100644
--- a/deps/pip/requirements_ncore.in
+++ b/deps/pip/requirements_ncore.in
@@ -24,10 +24,11 @@ pillow>=12.2.0; python_version>="3.9"
 pillow; python_version<"3.9"
 
 # zarr version rationales:
-# - 2.17.0 contains crucial perf improvements ["Cache result of FSStore._fsspec_installed()" https://github.com/zarr-developers/zarr-python/pull/1581]
+# - both zarr-python 2 and zarr-python 3 are supported
 # - 2.12.0 is the last version supporting python3.8 with old numpy versions, which we still want to support for now
-# - we can't switch to zarr 3 yet unconditionally as it's not available for python3.8 anymore
-zarr>=2.12.0,<3.0.0
+# - 2.17.0 contains crucial perf improvements ["Cache result of FSStore._fsspec_installed()" https://github.com/zarr-developers/zarr-python/pull/1581]
+# - 3.1.4 is the oldest supported zarr-python 3 version (Array.async_array, Store ABC without set_partial_values)
+zarr>=2.12.0,!=3.0.*,!=3.1.0,!=3.1.1,!=3.1.2,!=3.1.3
 
 # cbor2 version rationale:
 # - 5.9.0 fixes CWE-674 / CVE-2026-26209 (uncontrolled recursion DoS)
diff --git a/docs/data/formats.rst b/docs/data/formats.rst
index 125612f7..86d32841 100644
--- a/docs/data/formats.rst
+++ b/docs/data/formats.rst
@@ -107,7 +107,7 @@ might attach raw GPS/IMU measurements to the poses component:
        ├── {component_specific_data}...
        │
        └── generic_data/           (optional, only if generic data was set)
-           ├── {dataset_name}      [shape] dtype  (lz4-compressed zarr dataset)
+           ├── {array_name}        [shape] dtype  (lz4-compressed array)
            └── ...
 
 Writers use :meth:`~ncore.impl.data.v4.components.ComponentWriter.set_generic_data`
@@ -414,12 +414,12 @@ Supported categories:
 Encoding
 ^^^^^^^^
 
-* ``RAW`` -- Numpy array stored as a zarr dataset regular compression. Shape
+* ``RAW`` -- Numpy array stored as an array with regular compression. Shape
   is ``[H, W] + shape_suffix`` (e.g., ``[H, W, 2]`` for optical flow).
   Transparent quantization of raw labels is supported optionally
   (e.g., float32 depth quantized to uint16 with scale/offset).
 * ``IMAGE_ENCODED`` -- Pre-encoded image bytes (PNG, JPEG) stored as a 1-D
-  zarr uint8 dataset with no compression. Consumers can call ``get_encoded_data()`` for raw
+  uint8 array with no compression. Consumers can call ``get_encoded_data()`` for raw
   bytes (GPU-based decoding) or ``get_data()`` for Pillow-decoded numpy arrays.
 
 Instance naming convention
@@ -486,7 +486,7 @@ A minimal custom component looks like:
                ...  # collect data
 
            def finalize(self):
-               ...  # write zarr datasets to self._group
+               ...  # write arrays to self._group, e.g. self._group.create_array("velocities", ...)
 
        class Reader(ComponentReader):
            @staticmethod
@@ -498,7 +498,16 @@ A minimal custom component looks like:
                return version == "v1"
 
            def get_velocities(self):
-               return self._group["velocities"][:], self._group["timestamps_us"][:]
+               return (
+                   self._group.array("velocities").read(),
+                   self._group.array("timestamps_us").read(),
+               )
+
+The groups of component writers / readers (``self._group``) are
+``ncore.impl.data.nodes.Group`` instances with a typed API to create and read groups, arrays
+(``create_array`` / ``array(path).read()``), binary blobs (``create_bytes_array`` /
+``read_bytes``), and attributes. The wrapped zarr-python group is accessible via
+``self._group.zarr``.
 
 Writers must ensure that all stored timestamps fall within the sequence's
 ``sequence_timestamp_interval_us`` time range. Existing datasets can be extended
diff --git a/docs/data/storage_and_access.rst b/docs/data/storage_and_access.rst
index 91df6110..1dee9f62 100644
--- a/docs/data/storage_and_access.rst
+++ b/docs/data/storage_and_access.rst
@@ -34,8 +34,8 @@ capability.
    appends a compressed index enabling O(1) key lookups and direct seeks to any
    chunk.
 
-The ``.itar`` store implements the abstract zarr ``Store`` interface, so it can
-be used as a drop-in replacement for directory stores in all NCore APIs. Via
+The ``.itar`` store can be used as a drop-in replacement for directory stores in
+all NCore APIs. Via
 `UPath <https://github.com/fsspec/universal_pathlib>`_, ``.itar`` containers can
 also be accessed transparently from cloud storage backends (e.g., S3, GCS)
 without requiring a local copy.
@@ -137,6 +137,87 @@ random access without any index.
 .. |times| unicode:: U+00D7
 .. |rarr| unicode:: U+2192
 
+.. _zarr_versions:
+
+zarr Versions and Formats
+-------------------------
+
+NCore supports both major versions of the `zarr-python <https://zarr.readthedocs.io/>`_
+library, as well as both versions of the zarr on-disk format specification:
+
+* **zarr-python** (the installed library): ``2.x`` (Python |ge| 3.8) or ``3.x``
+  (Python |ge| 3.11, version |ge| 3.1.4). The matching implementation is selected
+  automatically at import time.
+* **zarr format** (the on-disk layout of component stores): format v2 (``.zgroup`` /
+  ``.zarray`` / ``.zattrs`` node metadata) or format v3 (``zarr.json`` node metadata).
+
+.. list-table::
+   :header-rows: 1
+   :widths: 30 35 35
+
+   * -
+     - zarr-python 2
+     - zarr-python 3
+   * - read zarr format v2
+     - |check|
+     - |check|
+   * - read zarr format v3
+     - |mdash|
+     - |check|
+   * - write zarr format v2 (default)
+     - |check|
+     - |check|
+   * - write zarr format v3
+     - |mdash|
+     - |check| (opt-in)
+
+zarr format v2 data written with either zarr-python version is equivalent (chunk records
+are byte-identical, node metadata documents are equivalent JSON) and can be read by all
+NCore V4 versions (NCore |le| 19.8 supports zarr-python 2 and zarr format v2 only). Data is
+therefore written in zarr format v2 by default. zarr format v3 can be selected explicitly
+when writing, e.g.
+
+.. code-block:: python
+
+   writer = SequenceComponentGroupsWriter(..., zarr_format=3)
+
+or via ``--zarr-format 3`` in the data converter CLIs. In zarr format v3, binary blobs
+(e.g., encoded images) are stored as ``uint8`` arrays (zarr format v2: fixed-length bytes).
+
+Both formats store the consolidated metadata of each component group as a single
+compressed record (``.zmetadata.cbor.xz`` for zarr format v2, CBOR + LZMA encoded;
+``zarr.json.cbor.zlib`` for zarr format v3, the CBOR + zlib encoded root ``zarr.json``
+including the consolidated metadata of the hierarchy, which is only stored in this
+compressed form). Hierarchies opened with consolidated metadata (the default of NCore
+readers) resolve all zarr nodes from it, without reading node metadata records. With
+zarr-python 3, NCore executes zarr's asynchronous I/O on the calling thread, avoiding
+per-access thread hand-offs to zarr-python's global I/O thread.
+
+With zarr-python 3, numeric arrays of ``.itar`` archives (blosc compressed or
+uncompressed, as written by NCore) are read by decompressing their chunks directly into
+the output array (zarr-python's indexing determines the chunks), and large chunks of an
+array are decoded concurrently. All other arrays and selections are read via
+zarr-python's codec pipeline. Setting the environment variable
+``NCORE_ZARR3_DIRECT_READS=0`` reads all arrays via zarr-python's codec pipeline.
+
+Records of ``.itar`` archives are write-once (archives are append-only): rewriting a
+record fails. zarr-python 3 rewrites node metadata on every attribute update, so node
+metadata records are kept in memory until the archive is closed, and only their final
+versions are appended (parents before children).
+
+.. note::
+
+   The groups of component readers / writers (``self._group``) are
+   ``ncore.impl.data.nodes.Group`` instances, which support both zarr-python
+   versions and zarr formats with a typed API (e.g. ``create_array``,
+   ``array(path).read()``, ``create_bytes_array`` / ``read_bytes``, ``attrs``). The
+   wrapped zarr-python objects are accessible via ``.zarr`` for components implemented
+   with the zarr-python API of the installed version directly.
+
+.. |check| unicode:: U+2713
+.. |ge| unicode:: U+2265
+.. |le| unicode:: U+2264
+
 Loading V4 Data
 ---------------
 
@@ -157,6 +238,50 @@ V4 sequences are loaded by specifying one or more component store paths:
    poses_readers = reader.open_component_readers(PosesComponent.Reader)
    camera_readers = reader.open_component_readers(CameraSensorComponent.Reader)
 
+Asynchronous Reads
+~~~~~~~~~~~~~~~~~~
+
+All component reader methods reading array data have asynchronous variants
+(``*_async``, e.g. ``get_frame_data_async``, ``get_frame_ray_bundle_data_async``,
+``get_generic_data_async``), returning the same data. Awaiting multiple reads
+concurrently overlaps their I/O and decompression (e.g., range requests to remote
+storage), and the running event loop isn't blocked (blocking work, such as
+decompression, runs on worker threads):
+
+.. code-block:: python
+
+   import asyncio
+
+   lidar = reader.open_component_readers(LidarSensorComponent.Reader)["lidar"]
+
+   async def load_frames(timestamps_us):
+       return await asyncio.gather(
+           *(lidar.get_frame_ray_bundle_data_async(ts, "direction") for ts in timestamps_us)
+       )
+
+   directions = asyncio.run(load_frames(lidar.frames_timestamps_us[:, 1].tolist()))
+
+Metadata (attributes, names, timestamps) is held in memory and only provided
+synchronously. Custom components can implement asynchronous accessors via
+``self._group.array(path).read_async()``.
+
+.. note::
+
+   Each in-flight read holds its decompressed data in memory. When reading many large
+   frames, bound the number of concurrent reads (e.g., to the number of CPU cores, or
+   higher for high-latency remote storage) instead of gathering all of them at once:
+
+   .. code-block:: python
+
+      async def load_frames(timestamps_us, max_concurrent_reads=16):
+          semaphore = asyncio.Semaphore(max_concurrent_reads)
+
+          async def load(ts):
+              async with semaphore:
+                  return await lidar.get_frame_ray_bundle_data_async(ts, "direction")
+
+          return await asyncio.gather(*(load(ts) for ts in timestamps_us))
+
 .. _cloud_storage_access:
 
 Cloud and Remote Storage Access
@@ -361,6 +486,14 @@ Performance Recommendations
   This is especially important for remote stores where each metadata
   lookup would otherwise be a separate round-trip.
 
+* **Keep the node cache enabled** (the default). Component readers cache the
+  groups / arrays they open (including their decoded metadata, but not their
+  data), so repeated accesses (e.g., of frames) skip re-opening them. The
+  ``node_cache_size`` parameter on
+  :class:`~ncore.data.v4.SequenceComponentGroupsReader` bounds the number of
+  cached nodes per component reader for very long sequences (``None`` caches
+  all accessed nodes, ``0`` disables caching).
+
 * **Tune the block size and cache type** for your workload.  The
   ``default_block_size`` and ``default_cache_type`` parameters on ``UPath``
   control how much data is fetched per HTTP request and how it is cached
diff --git a/ncore/BUILD.bazel b/ncore/BUILD.bazel
index e71fe1f7..e1a5b44f 100644
--- a/ncore/BUILD.bazel
+++ b/ncore/BUILD.bazel
@@ -56,10 +56,11 @@ py_wheel(
         "dataclasses_json>=0.2.12",
         "pillow",
         # zarr version restrictions / suggestions rationale:
+        # - both zarr-python 2 and zarr-python 3 are supported
         # - 2.17.0 contains crucial perf improvements ["Cache result of FSStore._fsspec_installed()" https://github.com/zarr-developers/zarr-python/pull/1581]
         # - 2.12.0 is the last version supporting python3.8 with old numpy versions, which we still want to support for now
-        # - we can't switch to zarr 3 yet unconditionally as it's not available for python3.8 anymore
-        "zarr>=2.12.0,<3.0.0",
+        # - 3.1.4 is the oldest supported zarr-python 3 version (Array.async_array, Store ABC without set_partial_values)
+        "zarr>=2.12.0,!=3.0.*,!=3.1.0,!=3.1.1,!=3.1.2,!=3.1.3",
         "cbor2",
         "scipy",
         "torch",
diff --git a/ncore/impl/data/BUILD.bazel b/ncore/impl/data/BUILD.bazel
index 5d7237a7..c7206756 100644
--- a/ncore/impl/data/BUILD.bazel
+++ b/ncore/impl/data/BUILD.bazel
@@ -25,6 +25,7 @@ py_library(
         "types.py",
     ],
     deps = [
+        ":pylib_json",
         ":pylib_util",
         "//ncore/impl/common:pylib_transformations",
         requirement("dataclasses_json"),
@@ -33,13 +34,30 @@ py_library(
     ],
 )
 
+py_library(
+    name = "pylib_json",
+    srcs = [
+        "_json.py",
+    ],
+)
+
 py_library(
     name = "pylib_stores",
     srcs = [
+        "_backend.py",
+        "_zarr2.py",
+        "_zarr3.py",
+        "encoding.py",
+        "itar.py",
+        "nodes.py",
         "stores.py",
     ],
     deps = [
+        ":pylib_json",
         requirement("cbor2"),
+        requirement("numcodecs"),
+        requirement("numpy"),
+        requirement("typing_extensions"),
         requirement("universal_pathlib"),
         requirement("zarr"),
     ],
diff --git a/ncore/impl/data/_backend.py b/ncore/impl/data/_backend.py
new file mode 100644
index 00000000..88cc4658
--- /dev/null
+++ b/ncore/impl/data/_backend.py
@@ -0,0 +1,244 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Interface of the zarr-python version specific backends of :mod:`ncore.impl.data.nodes` / :mod:`.stores`, and
+types shared by the backends"""
+
+from __future__ import annotations
+
+import dataclasses
+import sys
+import typing
+
+from typing import TYPE_CHECKING, Any, Dict, List, Literal, Mapping, Optional, Tuple, Union, cast
+
+import numcodecs
+import zarr
+
+from ncore.impl.data.itar import IndexedTarStore
+
+
+if sys.version_info >= (3, 10):
+    from types import EllipsisType
+else:  # pragma: no cover
+    EllipsisType = type(Ellipsis)
+
+if TYPE_CHECKING:
+    import numpy as np
+    import numpy.typing as npt  # type: ignore[import-not-found]
+
+    from typing_extensions import Protocol
+else:
+    Protocol = object
+
+
+#: Key of the compressed consolidated metadata of zarr format v2 hierarchies (CBOR + LZMA encoded ``.zmetadata``)
+V2_CONSOLIDATED_METADATA_KEY = ".zmetadata.cbor.xz"
+
+#: Key of the compressed consolidated metadata of zarr format v3 hierarchies (CBOR + zlib encoded root ``zarr.json``
+#: including the consolidated metadata of the hierarchy, which decodes several times faster than CBOR + LZMA). The root
+#: ``zarr.json`` of such hierarchies is only stored in this compressed form.
+V3_CONSOLIDATED_METADATA_KEY = "zarr.json.cbor.zlib"
+
+#: Basic selections of array data (integers, slices and Ellipsis)
+Selection = Union[int, slice, EllipsisType, Tuple[Union[int, slice, EllipsisType], ...]]
+
+#: Attributes of nodes (JSON-serializable values, also nested mappings / sequences: values are JSON-encoded by zarr-python,
+#: see :data:`ncore.impl.data.types.JsonLike` for the attributes read back)
+Attributes = Mapping[str, object]
+
+
+def attributes_dict(attributes: Attributes) -> Dict[str, Any]:
+    """A (shallow) copy of attributes as dictionary, as accepted by zarr-python"""
+    return dict(attributes)
+
+
+class ZarrStore:
+    """A zarr-python store of the installed zarr-python version (e.g., of a directory)"""
+
+
+#: Stores of zarr hierarchies: indexed tar archives, or zarr-python stores
+StoreLike = Union[IndexedTarStore, ZarrStore]
+
+
+def is_v2_metadata_key(key: str) -> bool:
+    """True for keys of zarr format v2 node metadata"""
+    return key.endswith((".zarray", ".zgroup", ".zattrs"))
+
+
+#: Blosc compressors supported by zarr-python 2 (numcodecs ``Blosc``) and zarr-python 3 (``BloscCodec``)
+BloscCname = Literal["blosclz", "lz4", "lz4hc", "zlib", "zstd"]
+
+#: Blosc shuffle modes (numcodecs ``Blosc`` encodes them as 0 / 1 / 2)
+BloscShuffle = Literal["noshuffle", "shuffle", "bitshuffle"]
+
+_BLOSC_SHUFFLES: List[BloscShuffle] = ["noshuffle", "shuffle", "bitshuffle"]
+
+
+@dataclasses.dataclass(frozen=True)
+class BloscCompression:
+    """Blosc compression of array chunks"""
+
+    #: Compressor used by blosc
+    cname: BloscCname = "lz4"
+    #: Compression level (0: no compression, 9: maximum compression)
+    clevel: int = 5
+    #: Shuffle filter applied before compression
+    shuffle: BloscShuffle = "bitshuffle"
+    #: Block size in bytes (0: automatic)
+    blocksize: int = 0
+
+    def __post_init__(self) -> None:
+        if self.cname not in typing.get_args(BloscCname):
+            raise ValueError(f"Unsupported blosc compressor {self.cname!r}, supported: {typing.get_args(BloscCname)}")
+        if not 0 <= self.clevel <= 9:
+            raise ValueError(f"Blosc compression level needs to be in [0, 9], got {self.clevel}")
+        if self.shuffle not in _BLOSC_SHUFFLES:
+            raise ValueError(f"Unsupported blosc shuffle {self.shuffle!r}, supported: {_BLOSC_SHUFFLES}")
+        if self.blocksize < 0:
+            raise ValueError(f"Blosc block size needs to be non-negative, got {self.blocksize}")
+
+    def to_config(self) -> Dict[str, object]:
+        """The configuration of the zarr format v3 ``blosc`` codec"""
+        return {"cname": self.cname, "clevel": self.clevel, "shuffle": self.shuffle, "blocksize": self.blocksize}
+
+    @staticmethod
+    def from_config(config: Mapping[str, object]) -> BloscCompression:
+        """From the configuration of a zarr format v3 ``blosc`` codec"""
+        return BloscCompression(
+            cname=cast(BloscCname, config["cname"]),
+            clevel=cast(int, config["clevel"]),
+            shuffle=cast(BloscShuffle, config["shuffle"]),
+            blocksize=cast(int, config.get("blocksize", 0)),
+        )
+
+    def to_numcodecs(self) -> numcodecs.Blosc:
+        return numcodecs.Blosc(
+            cname=self.cname,
+            clevel=self.clevel,
+            shuffle=_BLOSC_SHUFFLES.index(self.shuffle),
+            blocksize=self.blocksize,
+        )
+
+    @staticmethod
+    def from_numcodecs(compressor: numcodecs.Blosc) -> BloscCompression:
+        config = compressor.get_config()
+        return BloscCompression(
+            cname=config["cname"],
+            clevel=config["clevel"],
+            shuffle=_BLOSC_SHUFFLES[config["shuffle"]],
+            blocksize=config["blocksize"],
+        )
+
+
+ZarrNode = Union[zarr.Group, zarr.Array]
+
+
+class Backend(Protocol):
+    """zarr-python version specific operations of nodes and stores"""
+
+    #: zarr formats supported by the zarr-python version
+    supported_zarr_formats: Tuple[int, ...]
+
+    # -- nodes -----------------------------------------------------------------
+
+    def zarr_format(self, node: ZarrNode) -> int:
+        """The zarr format of a node"""
+        ...
+
+    def open_member(self, group: zarr.Group, path: str) -> ZarrNode:
+        """Opens a (nested) member of a group (raises KeyError if it doesn't exist)"""
+        ...
+
+    def member_names(self, group: zarr.Group) -> List[str]:
+        """Names of the direct members of a group"""
+        ...
+
+    def has_member(self, group: zarr.Group, name: str) -> bool:
+        """True if a group has a direct member with the given name"""
+        ...
+
+    def member_groups(self, group: zarr.Group) -> List[Tuple[str, zarr.Group]]:
+        """Direct member groups of a group"""
+        ...
+
+    def create_group(self, group: zarr.Group, name: str, attributes: Optional[Attributes]) -> zarr.Group:
+        """Creates a member group (raises if a member with the name exists already)"""
+        ...
+
+    def require_group(self, group: zarr.Group, name: str) -> zarr.Group:
+        """Returns a member group, creating it if it doesn't exist yet"""
+        ...
+
+    def create_array(
+        self,
+        group: zarr.Group,
+        name: str,
+        data: npt.NDArray[np.generic],
+        chunks: Union[Tuple[int, ...], Literal["auto"]],
+        compression: Optional[BloscCompression],
+        attributes: Optional[Attributes],
+    ) -> zarr.Array:
+        """Creates a member array initialized with data"""
+        ...
+
+    def read(self, array: zarr.Array, selection: Selection) -> object:
+        """Reads (a selection of) the data of an array (0-dimensional selections may result in numpy scalars)"""
+        ...
+
+    async def read_async(self, array: zarr.Array, selection: Selection) -> object:
+        """Reads (a selection of) the data of an array asynchronously, without blocking the running event loop"""
+        ...
+
+    def compression(self, array: zarr.Array) -> Optional[BloscCompression]:
+        """The blosc compression of an array (None if uncompressed, raises ValueError for other compressors)"""
+        ...
+
+    # -- stores ----------------------------------------------------------------
+
+    def open_directory_store(self, path: str) -> ZarrStore:
+        """Opens a directory store (a directory per group / array, a file per record)"""
+        ...
+
+    def create_root_group(self, store: StoreLike, zarr_format: Literal[2, 3], attributes: Attributes) -> zarr.Group:
+        """Creates the root group of a hierarchy in a store"""
+        ...
+
+    def open_store(self, store: StoreLike, consolidated: bool) -> zarr.Group:
+        """Opens the root group of a hierarchy for reading, via its compressed consolidated metadata if consolidated
+        (raises KeyError if there is no compressed consolidated metadata)"""
+        ...
+
+    def consolidate_store(self, store: StoreLike) -> None:
+        """Stores the compressed consolidated metadata of the hierarchy of a store"""
+        ...
+
+    def group_store(self, group: zarr.Group) -> StoreLike:
+        """The store of a group (the indexed tar store for groups of indexed tar archives)"""
+        ...
+
+
+__all__ = [
+    "Attributes",
+    "attributes_dict",
+    "Backend",
+    "BloscCompression",
+    "Selection",
+    "StoreLike",
+    "V2_CONSOLIDATED_METADATA_KEY",
+    "V3_CONSOLIDATED_METADATA_KEY",
+    "ZarrStore",
+    "is_v2_metadata_key",
+]
diff --git a/ncore/impl/data/_json.py b/ncore/impl/data/_json.py
new file mode 100644
index 00000000..d19174bf
--- /dev/null
+++ b/ncore/impl/data/_json.py
@@ -0,0 +1,35 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""JSON-like values (independent of the other data modules)"""
+
+from __future__ import annotations
+
+from typing import Dict, List, Union
+
+
+#: A JSON-serializable value
+JsonLike = Union[
+    Dict[str, "JsonLike"],
+    List["JsonLike"],
+    str,
+    int,
+    float,
+    bool,
+    None,
+]
+
+#: A JSON-serializable document (a dictionary at the top level)
+JsonDocument = Dict[str, JsonLike]
diff --git a/ncore/impl/data/_zarr2.py b/ncore/impl/data/_zarr2.py
new file mode 100644
index 00000000..181a99a9
--- /dev/null
+++ b/ncore/impl/data/_zarr2.py
@@ -0,0 +1,210 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""zarr-python 2 backend of :mod:`ncore.impl.data.nodes` and :mod:`ncore.impl.data.stores`"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+
+from typing import TYPE_CHECKING, Any, Iterator, List, Literal, Optional, Tuple, Union
+
+import numcodecs
+import numpy as np
+import zarr
+
+from numcodecs.compat import ensure_bytes
+
+from ncore.impl.data._backend import (
+    V2_CONSOLIDATED_METADATA_KEY,
+    Attributes,
+    BloscCompression,
+    Selection,
+    StoreLike,
+    ZarrStore,
+    attributes_dict,
+    is_v2_metadata_key,
+)
+from ncore.impl.data.encoding import decode_cbor_xz, encode_cbor_xz
+from ncore.impl.data.itar import IndexedTarStore
+
+
+if TYPE_CHECKING:
+    import numpy.typing as npt  # type: ignore[import-not-found]
+
+    _StoreBase = object
+    _ConsolidatedMetadataStoreBase = object
+else:
+    from zarr._storage.store import Store as _StoreBase
+    from zarr.storage import ConsolidatedMetadataStore as _ConsolidatedMetadataStoreBase
+
+
+#: zarr-python 2 API (untyped, type checking may run against zarr-python 3)
+_zarr: Any = zarr
+
+
+def _z(node: object) -> Any:
+    """A zarr-python 2 group / array / store (untyped, see :data:`_zarr`)"""
+    return node
+
+
+class _TarStore(_StoreBase):
+    """zarr-python 2 store of the records of an :class:`IndexedTarStore`"""
+
+    _erasable = False
+
+    def __init__(self, store: IndexedTarStore) -> None:
+        self.store = store
+
+    def __getitem__(self, key: str) -> bytes:
+        return self.store[key]
+
+    def __setitem__(self, key: str, value: object) -> None:
+        if self.store.mode != "w":
+            raise _zarr.errors.ReadOnlyError()
+        self.store[key] = ensure_bytes(value)
+
+    def __delitem__(self, key: str) -> None:
+        raise NotImplementedError("IndexedTarStore: records can't be deleted")
+
+    def __contains__(self, key: object) -> bool:
+        return key in self.store
+
+    def __iter__(self) -> Iterator[str]:
+        return iter(self.store.keys())
+
+    def __len__(self) -> int:
+        return len(self.store)
+
+    def __eq__(self, value: object) -> bool:
+        return isinstance(value, _TarStore) and self.store == value.store
+
+    def __hash__(self) -> int:
+        return hash(self.store)
+
+    def close(self) -> None:
+        self.store.close()
+
+
+class _CompressedConsolidatedStore(_ConsolidatedMetadataStoreBase):
+    """A store serving all node metadata from the compressed consolidated metadata of the wrapped store"""
+
+    def __init__(self, store: Any) -> None:
+        self.store = store
+        consolidated = decode_cbor_xz(store[V2_CONSOLIDATED_METADATA_KEY])
+        if (consolidated_format := consolidated.get("zarr_consolidated_format")) != 1:
+            raise _zarr.errors.MetadataError(f"unsupported zarr consolidated metadata format: {consolidated_format}")
+        self.meta_store = _zarr.storage.KVStore(consolidated["metadata"])
+
+
+def _zarr_store(store: StoreLike) -> Any:
+    """The zarr-python 2 store of a store (wraps indexed tar stores)"""
+    return _TarStore(store) if isinstance(store, IndexedTarStore) else store
+
+
+class Backend:
+    """zarr-python 2 implementation of the version specific operations of nodes and stores"""
+
+    supported_zarr_formats: Tuple[int, ...] = (2,)
+
+    # -- nodes -----------------------------------------------------------------
+
+    def zarr_format(self, node: Union[zarr.Group, zarr.Array]) -> int:
+        return 2
+
+    def open_member(self, group: zarr.Group, path: str) -> Union[zarr.Group, zarr.Array]:
+        return _z(group)[path]
+
+    def member_names(self, group: zarr.Group) -> List[str]:
+        return list(_z(group).keys())
+
+    def has_member(self, group: zarr.Group, name: str) -> bool:
+        return name in _z(group)
+
+    def member_groups(self, group: zarr.Group) -> List[Tuple[str, zarr.Group]]:
+        return list(_z(group).groups())
+
+    def create_group(self, group: zarr.Group, name: str, attributes: Optional[Attributes]) -> zarr.Group:
+        child = _z(group).create_group(name)
+        if attributes is not None:
+            child.attrs.update(attributes_dict(attributes))  # (writes .zattrs once)
+        return child
+
+    def require_group(self, group: zarr.Group, name: str) -> zarr.Group:
+        return _z(group).require_group(name)
+
+    def create_array(
+        self,
+        group: zarr.Group,
+        name: str,
+        data: npt.NDArray[np.generic],
+        chunks: Union[Tuple[int, ...], Literal["auto"]],
+        compression: Optional[BloscCompression],
+        attributes: Optional[Attributes],
+    ) -> zarr.Array:
+        array = _z(group).create_dataset(
+            name,
+            data=data,
+            chunks=True if chunks == "auto" else chunks,
+            compressor=None if compression is None else compression.to_numcodecs(),
+        )
+        if attributes is not None:
+            array.attrs.update(attributes_dict(attributes))
+        return array
+
+    def read(self, array: zarr.Array, selection: Selection) -> object:
+        return _z(array)[selection]
+
+    async def read_async(self, array: zarr.Array, selection: Selection) -> object:
+        # (zarr-python 2 is synchronous: reads run on worker threads, to not block the running event loop)
+        return await asyncio.get_running_loop().run_in_executor(None, self.read, array, selection)
+
+    def compression(self, array: zarr.Array) -> Optional[BloscCompression]:
+        if (compressor := _z(array).compressor) is None:
+            return None
+        if not isinstance(compressor, numcodecs.Blosc):
+            raise ValueError(f"Unsupported compressor {compressor}")
+        return BloscCompression.from_numcodecs(compressor)
+
+    # -- stores ----------------------------------------------------------------
+
+    def open_directory_store(self, path: str) -> ZarrStore:
+        return _zarr.storage.DirectoryStore(path)
+
+    def create_root_group(self, store: StoreLike, zarr_format: Literal[2, 3], attributes: Attributes) -> zarr.Group:
+        root = _zarr.group(store=_zarr_store(store))
+        root.attrs.put(attributes_dict(attributes))
+        return root
+
+    def open_store(self, store: StoreLike, consolidated: bool) -> zarr.Group:
+        zarr_store = _zarr_store(store)
+        if consolidated:
+            return _zarr.open_group(store=_CompressedConsolidatedStore(zarr_store), chunk_store=zarr_store, mode="r")
+        return _zarr.open_group(store=zarr_store, mode="r")
+
+    def consolidate_store(self, store: StoreLike) -> None:
+        zarr_store = _zarr_store(store)
+        metadata = {key: json.loads(zarr_store[key]) for key in zarr_store if is_v2_metadata_key(key)}
+        zarr_store[V2_CONSOLIDATED_METADATA_KEY] = encode_cbor_xz({"zarr_consolidated_format": 1, "metadata": metadata})
+
+    def group_store(self, group: zarr.Group) -> StoreLike:
+        store = _z(group).store
+        if isinstance(store, _CompressedConsolidatedStore):
+            store = store.store
+        return store.store if isinstance(store, _TarStore) else store
+
+
+backend = Backend()
diff --git a/ncore/impl/data/_zarr3.py b/ncore/impl/data/_zarr3.py
new file mode 100644
index 00000000..1edcb108
--- /dev/null
+++ b/ncore/impl/data/_zarr3.py
@@ -0,0 +1,756 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""zarr-python 3 backend of :mod:`ncore.impl.data.nodes` and :mod:`ncore.impl.data.stores` (only imported if
+zarr-python 3 is installed).
+
+zarr-python 3 executes all I/O as coroutines. Its synchronous API submits each call to a global event loop running on
+a dedicated I/O thread, and codecs additionally hand off (de)compression to worker threads. For many small accesses
+(as typical for reading individual frames) these thread hand-offs dominate. The backend instead runs zarr-python's
+coroutines on an event loop owned by the *calling* thread (see :func:`run`), which executes the same zarr-python code
+paths without thread hand-offs, and lets multiple reader threads decode data concurrently.
+
+Arrays of indexed tar archives are read by decompressing their chunks directly into the output array (see
+:func:`_direct_read`).
+"""
+
+from __future__ import annotations
+
+import asyncio
+import concurrent.futures
+import contextlib
+import json
+import os
+import sys
+import threading
+import warnings
+
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    AsyncIterator,
+    Callable,
+    Coroutine,
+    Dict,
+    Iterable,
+    Iterator,
+    List,
+    Literal,
+    NamedTuple,
+    Optional,
+    Tuple,
+    TypeVar,
+    Union,
+    cast,
+)
+
+import numcodecs
+import numpy as np
+import zarr
+import zarr.api.asynchronous as zarr_async
+
+from zarr.abc.codec import Codec
+from zarr.abc.store import ByteRequest, OffsetByteRequest, RangeByteRequest, Store, SuffixByteRequest
+from zarr.codecs import BloscCodec
+from zarr.core.buffer import Buffer, BufferPrototype, default_buffer_prototype
+from zarr.core.group import AsyncGroup
+from zarr.core.indexing import BasicIndexer
+from zarr.core.sync import sync
+from zarr.storage import LocalStore, StorePath
+
+from ncore.impl.data._backend import (
+    V2_CONSOLIDATED_METADATA_KEY,
+    V3_CONSOLIDATED_METADATA_KEY,
+    Attributes,
+    BloscCompression,
+    Selection,
+    StoreLike,
+    ZarrStore,
+    attributes_dict,
+    is_v2_metadata_key,
+)
+from ncore.impl.data.encoding import decode_cbor_xz, decode_cbor_zlib, encode_cbor_xz, encode_cbor_zlib
+from ncore.impl.data.itar import IndexedTarStore
+
+
+if TYPE_CHECKING:
+    import numpy.typing as npt  # type: ignore[import-not-found]
+
+
+T = TypeVar("T")
+
+
+# -----------------------------------------------------------------------------
+# Execution of coroutines on event loops of the calling threads
+# -----------------------------------------------------------------------------
+
+
+class _InlineExecutor(concurrent.futures.ThreadPoolExecutor):
+    """Executor running submitted work immediately in the submitting thread (asyncio requires default executors to be
+    thread pools, which start no threads until work is submitted to them)"""
+
+    def submit(self, fn, /, *args, **kwargs):  # type: ignore[override]
+        future: concurrent.futures.Future = concurrent.futures.Future()
+        try:
+            future.set_result(fn(*args, **kwargs))
+        except BaseException as e:  # forwarded to the awaiting coroutine
+            future.set_exception(e)
+        return future
+
+
+class _Loop:
+    """An event loop of a thread (closed when the thread ends)"""
+
+    def __init__(self) -> None:
+        self.loop = asyncio.new_event_loop()
+        self.loop.set_default_executor(_InlineExecutor())  # (codecs offload (de)compression to the default executor)
+        self.pid = os.getpid()
+
+    def __del__(self) -> None:
+        # Loops of parent processes aren't closed in forked processes: their selectors (e.g., epoll instances) are
+        # shared with the parent, closing them would unregister the parent's file descriptors (e.g., the self-pipe of
+        # the parent's loop, which then misses wake-ups by other threads)
+        if self.pid == os.getpid() and not self.loop.is_closed() and not self.loop.is_running():
+            self.loop.close()
+
+
+class _ThreadLoop(threading.local):
+    state: Optional[_Loop] = None
+
+
+_thread_loop = _ThreadLoop()
+
+
+def _on_own_loop() -> bool:
+    """True if running on the event loop of :func:`run` of the calling thread (blocking work runs inline), False on
+    other event loops (blocking work is offloaded to threads, to not block them)"""
+    state = _thread_loop.state
+    return state is not None and state.pid == os.getpid() and asyncio.get_running_loop() is state.loop
+
+
+def run(coroutine: Coroutine[Any, Any, T]) -> T:
+    """Runs a zarr-python coroutine to completion on an event loop owned by the calling thread.
+
+    The loop runs offloaded work (e.g., of codecs) inline. Processes forked by a thread with a loop create a new loop.
+    Within running event loops (e.g., of asynchronous applications), coroutines run on zarr-python's I/O loop.
+    """
+    try:
+        asyncio.get_running_loop()
+    except RuntimeError:
+        if (state := _thread_loop.state) is None or state.pid != os.getpid():
+            state = _thread_loop.state = _Loop()  # (see _Loop.__del__ for loops of parent processes)
+        return state.loop.run_until_complete(coroutine)
+    return sync(coroutine)
+
+
+# -----------------------------------------------------------------------------
+# zarr-python 3 store of indexed tar archives
+# -----------------------------------------------------------------------------
+
+
+def _byte_range(byte_range: Optional[ByteRequest], size: int) -> Tuple[int, int]:
+    """Maps a zarr-python byte request to the (start, end) range of a record of the given size"""
+    if byte_range is None:
+        return 0, size
+    if isinstance(byte_range, RangeByteRequest):
+        return min(byte_range.start, size), min(byte_range.end, size)
+    if isinstance(byte_range, OffsetByteRequest):
+        return min(byte_range.offset, size), size
+    if isinstance(byte_range, SuffixByteRequest):
+        return max(0, size - byte_range.suffix), size
+    raise TypeError(f"Unexpected byte range {byte_range}")
+
+
+class _TarStore(Store):
+    """zarr-python 3 store of the records of an :class:`IndexedTarStore` (synchronous I/O)"""
+
+    supports_writes: bool = True
+    supports_deletes: bool = False
+    supports_partial_writes: Literal[False] = False
+    supports_listing: bool = True
+
+    def __init__(self, store: IndexedTarStore) -> None:
+        super().__init__(read_only=store.mode == "r")
+        self.store = store
+        self._is_open = True
+
+    def __eq__(self, value: object) -> bool:
+        return isinstance(value, _TarStore) and self.store == value.store
+
+    def __hash__(self) -> int:
+        return hash(self.store)
+
+    def __repr__(self) -> str:
+        return f"_TarStore({self.store!r})"
+
+    def with_read_only(self, read_only: bool = False) -> _TarStore:
+        if read_only != self.read_only:
+            raise NotImplementedError("IndexedTarStore: the access mode of archives can't be changed")
+        return self
+
+    def close(self) -> None:
+        self.store.close()
+        self._is_open = False
+
+    def _get(
+        self, key: str, prototype: Optional[BufferPrototype], byte_range: Optional[ByteRequest]
+    ) -> Optional[Buffer]:
+        try:
+            value = self.store[key]
+        except KeyError:
+            return None
+        start, end = _byte_range(byte_range, len(value))
+        return (prototype or default_buffer_prototype()).buffer.from_bytes(value[start:end])
+
+    async def get(
+        self, key: str, prototype: BufferPrototype, byte_range: Optional[ByteRequest] = None
+    ) -> Optional[Buffer]:
+        return self._get(key, prototype, byte_range)
+
+    async def get_partial_values(
+        self, prototype: BufferPrototype, key_ranges: Iterable[Tuple[str, Optional[ByteRequest]]]
+    ) -> List[Optional[Buffer]]:
+        return [self._get(key, prototype, byte_range) for key, byte_range in key_ranges]
+
+    async def exists(self, key: str) -> bool:
+        return key in self.store
+
+    async def set(self, key: str, value: Buffer) -> None:
+        self._check_writable()
+        self.store[key] = value.to_bytes()
+
+    async def set_if_not_exists(self, key: str, value: Buffer) -> None:
+        if key not in self.store:
+            await self.set(key, value)
+
+    async def delete(self, key: str) -> None:
+        """No-op for records that don't exist (zarr-python deletes them when creating nodes)"""
+        self._check_writable()
+        if key in self.store:
+            raise NotImplementedError("IndexedTarStore: records can't be deleted")
+
+    async def delete_dir(self, prefix: str) -> None:
+        """No-op for prefixes without records (zarr-python deletes them when creating nodes)"""
+        self._check_writable()
+        prefix = prefix if prefix == "" or prefix.endswith("/") else prefix + "/"
+        if any(key.startswith(prefix) for key in self.store.keys()):
+            raise NotImplementedError("IndexedTarStore: records can't be deleted")
+
+    async def list(self) -> AsyncIterator[str]:
+        for key in self.store.keys():
+            yield key
+
+    async def list_prefix(self, prefix: str) -> AsyncIterator[str]:
+        for key in self.store.keys():
+            if key.startswith(prefix):
+                yield key
+
+    async def list_dir(self, prefix: str) -> AsyncIterator[str]:
+        prefix = prefix.rstrip("/")
+        prefix = prefix + "/" if prefix else ""
+        children = {key[len(prefix) :].split("/", 1)[0]: None for key in self.store.keys() if key.startswith(prefix)}
+        for child in children:
+            yield child
+
+
+def _zarr_store(store: StoreLike) -> Store:
+    """The zarr-python 3 store of a store (wraps indexed tar stores)"""
+    return _TarStore(store) if isinstance(store, IndexedTarStore) else cast(Store, store)
+
+
+def _keys(store: Store) -> List[str]:
+    async def keys() -> List[str]:
+        return [key async for key in store.list()]
+
+    return run(keys())
+
+
+def _get(store: Store, key: str) -> Optional[bytes]:
+    value = run(store.get(key, prototype=default_buffer_prototype()))
+    return None if value is None else value.to_bytes()
+
+
+def _parent_grouped(metadata: dict, node_path: Callable[[str], str]) -> dict:
+    """Orders consolidated node metadata by node depth, with the members of each group adjacent.
+
+    zarr-python < 3.4 nests flat consolidated metadata assuming this order (zarr-python#4226: the members of a group
+    are attached to the first run of keys of the group only), which isn't the order of hierarchies created in arbitrary
+    order (e.g., by multiple component writers). Keys are node metadata keys (zarr format v2) or node paths (zarr format
+    v3), ``node_path`` maps them to node paths.
+    """
+
+    def order(key: str) -> Tuple[int, str, str]:
+        parent, _, name = (path := node_path(key)).rpartition("/")
+        return path.count("/"), parent, name
+
+    return {key: metadata[key] for key in sorted(metadata, key=order)}
+
+
+def _v2_node_path(key: str) -> str:
+    return key.rpartition("/")[0]
+
+
+def _v3_node_documents(store: Store) -> Dict[str, object]:
+    """The node metadata documents of all members of a zarr format v3 hierarchy, by node path"""
+    documents: Dict[str, object] = {}
+    for key in _keys(store):
+        if key.endswith("/zarr.json") and (value := _get(store, key)) is not None:
+            documents[key[: -len("/zarr.json")]] = json.loads(value)
+    return documents
+
+
+@contextlib.contextmanager
+def _ignored_warnings() -> Iterator[None]:
+    with warnings.catch_warnings():
+        # compressed consolidated metadata records are not zarr nodes
+        warnings.filterwarnings("ignore", message="Object at .* is not recognized as a component")
+        yield
+
+
+# -----------------------------------------------------------------------------
+# Direct chunk reads of indexed tar archives
+# -----------------------------------------------------------------------------
+
+#: Environment variable disabling direct reads (all reads via zarr-python's codec pipeline), e.g., to compare against
+#: or to benefit from zarr-python improvements
+DIRECT_READS_ENV = "NCORE_ZARR3_DIRECT_READS"
+
+#: Minimum number of decompressed bytes per chunk to decode the chunks of a read concurrently (smaller chunks are
+#: cheaper to decode than to hand off to a thread pool)
+_PARALLEL_CHUNK_BYTES = 2 << 20
+
+#: Maximum number of concurrent reads (e.g., of multiple reader threads) decoding their chunks concurrently, further
+#: concurrent reads decode their chunks sequentially (the reads themselves already utilize the cores)
+_PARALLEL_READS = 2
+
+
+class _ParallelReads:
+    """Bounds the number of reads decoding their chunks concurrently (per process)"""
+
+    def __init__(self) -> None:
+        self.semaphore = threading.BoundedSemaphore(_PARALLEL_READS)
+        self.executor: Optional[concurrent.futures.ThreadPoolExecutor] = None
+        self.lock = threading.Lock()
+        self.pid = os.getpid()
+
+    def pool(self) -> concurrent.futures.ThreadPoolExecutor:
+        with self.lock:
+            if self.executor is None:
+                self.executor = concurrent.futures.ThreadPoolExecutor(
+                    max_workers=min(8, os.cpu_count() or 1), thread_name_prefix="ncore_zarr3_chunks"
+                )
+            return self.executor
+
+
+_parallel_reads = _ParallelReads()
+
+
+def _parallel() -> _ParallelReads:
+    """The parallel reads state of this process (re-created in forked processes, whose locks / threads are unusable)"""
+    global _parallel_reads
+    if _parallel_reads.pid != os.getpid():
+        _parallel_reads = _ParallelReads()
+    return _parallel_reads
+
+
+class _DirectRead(NamedTuple):
+    """How to read the chunks of an array directly (see :func:`_direct_read`)"""
+
+    store: IndexedTarStore
+    path: str
+    dtype: np.dtype
+    chunks: Tuple[int, ...]
+    chunk_key: Callable[[Tuple[int, ...]], str]
+    compressor: Optional[numcodecs.Blosc]
+    fill_value: object
+
+
+def _direct_reads_enabled() -> bool:
+    return os.environ.get(DIRECT_READS_ENV, "1").strip().lower() not in ("0", "false", "no", "off")
+
+
+def _direct_read(array: zarr.Array) -> Optional[_DirectRead]:
+    """How to read the chunks of an array directly (None if the array is read via zarr-python).
+
+    Supported are arrays of indexed tar archives opened for reading with a regular chunk grid, C order, numeric /
+    boolean data of native byte order or fixed-length bytes, no filters, and no compression or a single blosc compressor
+    (all arrays written by ncore). zarr-python determines the chunks and their selections (its indexers), but the
+    chunks are read and decompressed directly into the output array, instead of via zarr-python's codec pipeline, which
+    decodes each chunk into a temporary buffer and copies it (see zarr-python#2904).
+
+    Planned per read (sub-microsecond), so nothing outlives the array.
+    """
+    store_path = array.async_array.store_path
+    if not isinstance(store := store_path.store, _TarStore) or store.store.mode != "r":
+        return None
+
+    metadata: Any = array.metadata
+    dtype = np.dtype(metadata.dtype.to_native_dtype())
+    if dtype.hasobject or not dtype.isnative or dtype.kind not in "biufS":
+        return None
+    if metadata.zarr_format == 2:
+        if metadata.order != "C" or metadata.filters:
+            return None
+        compressor = metadata.compressor
+        chunks = tuple(metadata.chunks)
+    else:
+        codecs = list(metadata.codecs)
+        if not codecs or type(codecs[0]).__name__ != "BytesCodec":
+            return None
+        if dtype.itemsize > 1 and str(getattr(codecs[0].endian, "value", codecs[0].endian)) != sys.byteorder:
+            return None
+        if len(codecs) == 1:
+            compressor = None
+        elif len(codecs) == 2 and isinstance(codecs[1], BloscCodec):
+            compressor = codecs[1]._blosc_codec
+        else:
+            return None
+        if not hasattr(metadata.chunk_grid, "chunk_shape"):  # (regular chunk grids only)
+            return None
+        chunks = tuple(metadata.chunk_grid.chunk_shape)
+    if compressor is not None and not isinstance(compressor, numcodecs.Blosc):
+        return None
+    return _DirectRead(
+        store.store, store_path.path, dtype, chunks, metadata.encode_chunk_key, compressor, metadata.fill_value
+    )
+
+
+def _is_basic_selection(selection: Selection) -> bool:
+    """True for selections of (tuples of) integers, slices with unit step, and Ellipsis"""
+    for item in selection if isinstance(selection, tuple) else (selection,):
+        if item is Ellipsis or (isinstance(item, (int, np.integer)) and not isinstance(item, (bool, np.bool_))):
+            continue
+        if isinstance(item, slice) and item.step in (None, 1):
+            continue
+        return False
+    return True
+
+
+def _chunk_grid(array: zarr.Array) -> Any:
+    """zarr-python's chunk grid of an array (an array attribute since zarr-python 3.2, metadata before)"""
+    if (grid := getattr(array.async_array, "_chunk_grid", None)) is not None:
+        return grid
+    return cast(Any, array.metadata).chunk_grid
+
+
+class _DirectReadOp:
+    """A direct read of a basic selection of an array: the chunks to read (see :meth:`keys`), and their decoding into
+    the output array (see :meth:`decode`)"""
+
+    def __init__(self, plan: _DirectRead, array: zarr.Array, selection: Selection) -> None:
+        self.plan = plan
+        self.shape = tuple(array.metadata.shape)
+        indexer = BasicIndexer(selection, self.shape, _chunk_grid(array))
+        self.out_shape = tuple(indexer.shape)
+        self.projections: List[Any] = list(indexer) if all(indexer.shape) else []
+        self.keys = [self._key(projection) for projection in self.projections]
+
+    def _key(self, projection: Any) -> str:
+        key = self.plan.chunk_key(projection.chunk_coords)
+        return f"{self.plan.path}/{key}" if self.plan.path else key
+
+    @property
+    def is_single_raw_chunk(self) -> bool:
+        """True for selections of complete uncompressed single-chunk arrays (e.g., encoded images), whose chunk bytes
+        are the result"""
+        plan = self.plan
+        return (
+            plan.compressor is None
+            and len(self.projections) == 1
+            and self.projections[0].is_complete_chunk
+            and plan.chunks == self.shape == self.out_shape
+        )
+
+    @property
+    def decodes_concurrently(self) -> bool:
+        """True for reads of multiple large chunks (see :data:`_PARALLEL_CHUNK_BYTES`)"""
+        chunk_bytes = self.plan.dtype.itemsize * int(np.prod(self.plan.chunks))
+        return len(self.projections) > 1 and chunk_bytes >= _PARALLEL_CHUNK_BYTES
+
+    def decode(self, raws: List[Optional[bytes]], fetch: Optional[Callable[[str], Optional[bytes]]] = None) -> Any:
+        """Decodes the chunks (None: not stored, read as the fill value) into the output array. With ``fetch``, chunks
+        are read by the decoding threads (raws are ignored)"""
+        plan = self.plan
+        if self.is_single_raw_chunk and (raw := raws[0] if fetch is None else fetch(self.keys[0])) is not None:
+            # (bytearray: results are writeable, as results of reads via zarr-python)
+            return np.frombuffer(bytearray(raw), dtype=plan.dtype).reshape(self.shape)
+
+        out = np.empty(self.out_shape, dtype=plan.dtype)
+
+        def decode_chunk(index: int) -> None:
+            projection = self.projections[index]
+            raw = raws[index] if fetch is None else fetch(self.keys[index])
+            target = out[projection.out_selection] if projection.out_selection != () else out
+            if raw is None:  # chunk not stored (fill value)
+                target[...] = np.asarray(plan.fill_value, dtype=plan.dtype) if plan.fill_value is not None else 0
+                return
+            if (
+                plan.compressor is not None
+                and projection.is_complete_chunk
+                and target.shape == plan.chunks
+                and target.flags.c_contiguous
+            ):
+                plan.compressor.decode(raw, out=target)
+                return
+            decoded = plan.compressor.decode(raw) if plan.compressor is not None else raw
+            selected = np.frombuffer(decoded, dtype=plan.dtype).reshape(plan.chunks)[projection.chunk_selection]
+            target[...] = selected.reshape(target.shape)  # (integer selections drop axes)
+
+        parallel = _parallel()
+        if self.decodes_concurrently and parallel.semaphore.acquire(blocking=False):
+            try:
+                for future in [parallel.pool().submit(decode_chunk, i) for i in range(len(self.projections))]:
+                    future.result()
+            finally:
+                parallel.semaphore.release()
+        else:
+            for index in range(len(self.projections)):
+                decode_chunk(index)
+        return out  # (0-d arrays for 0-d results, see nodes.Array.read)
+
+
+def _get_record(store: IndexedTarStore, key: str) -> Optional[bytes]:
+    try:
+        return store[key]
+    except KeyError:
+        return None
+
+
+async def _get_record_async(store: IndexedTarStore, key: str) -> Optional[bytes]:
+    try:
+        return await store.read_async(key)
+    except KeyError:
+        return None
+
+
+def _read_direct(plan: _DirectRead, array: zarr.Array, selection: Selection) -> Any:
+    """Reads a basic selection of an array, decompressing its chunks directly into the output array.
+
+    Reads of multiple large chunks (see :data:`_PARALLEL_CHUNK_BYTES`) read and decode their chunks concurrently on a
+    thread pool (positional reads and blosc decompression release the GIL), unless :data:`_PARALLEL_READS` reads do so
+    already. All other reads run on the calling thread.
+    """
+    op = _DirectReadOp(plan, array, selection)
+    return op.decode([], fetch=lambda key: _get_record(plan.store, key))
+
+
+async def _read_direct_async(plan: _DirectRead, array: zarr.Array, selection: Selection) -> Any:
+    """Asynchronous :func:`_read_direct`: chunks of remote archives are read via concurrent asynchronous range
+    requests. Blocking work (local reads, decompression) runs on the calling thread on the event loop of :func:`run`,
+    on worker threads on other event loops (e.g., of asynchronous applications), so they aren't blocked."""
+    op = _DirectReadOp(plan, array, selection)
+    if plan.store.is_remote:
+        raws = list(await asyncio.gather(*(_get_record_async(plan.store, key) for key in op.keys)))
+        if _on_own_loop():
+            return op.decode(raws)
+        return await asyncio.get_running_loop().run_in_executor(_read_executor(), op.decode, raws)
+    if _on_own_loop():
+        return op.decode([], fetch=lambda key: _get_record(plan.store, key))
+    return await asyncio.get_running_loop().run_in_executor(
+        _read_executor(), lambda: op.decode([], fetch=lambda key: _get_record(plan.store, key))
+    )
+
+
+def _read_executor() -> concurrent.futures.ThreadPoolExecutor:
+    """Executor of blocking parts of reads awaited on event loops other than the one of :func:`run` (per process)"""
+    global _read_executor_state
+    if _read_executor_state is None or _read_executor_state[0] != os.getpid():
+        _read_executor_state = (
+            os.getpid(),
+            concurrent.futures.ThreadPoolExecutor(
+                max_workers=min(32, (os.cpu_count() or 1) + 4), thread_name_prefix="ncore_zarr3_reads"
+            ),
+        )
+    return _read_executor_state[1]
+
+
+_read_executor_state: Optional[Tuple[int, concurrent.futures.ThreadPoolExecutor]] = None
+
+
+# -----------------------------------------------------------------------------
+# Backend
+# -----------------------------------------------------------------------------
+
+
+class Backend:
+    """zarr-python 3 implementation of the version specific operations of nodes and stores"""
+
+    supported_zarr_formats: Tuple[int, ...] = (2, 3)
+
+    # -- nodes -----------------------------------------------------------------
+
+    def zarr_format(self, node: Union[zarr.Group, zarr.Array]) -> int:
+        return node.metadata.zarr_format
+
+    def open_member(self, group: zarr.Group, path: str) -> Union[zarr.Group, zarr.Array]:
+        node = run(group._async_group.getitem(path))
+        return zarr.Group(node) if isinstance(node, AsyncGroup) else zarr.Array(node)
+
+    def member_names(self, group: zarr.Group) -> List[str]:
+        with _ignored_warnings():  # (members of groups with consolidated metadata are listed without I/O)
+            return list(group.keys())
+
+    def has_member(self, group: zarr.Group, name: str) -> bool:
+        return name in self.member_names(group)
+
+    def member_groups(self, group: zarr.Group) -> List[Tuple[str, zarr.Group]]:
+        with _ignored_warnings():
+            return list(group.groups())
+
+    def create_group(self, group: zarr.Group, name: str, attributes: Optional[Attributes]) -> zarr.Group:
+        return group.create_group(name, attributes=None if attributes is None else attributes_dict(attributes))
+
+    def require_group(self, group: zarr.Group, name: str) -> zarr.Group:
+        return group.require_group(name)
+
+    def create_array(
+        self,
+        group: zarr.Group,
+        name: str,
+        data: npt.NDArray[np.generic],
+        chunks: Union[Tuple[int, ...], Literal["auto"]],
+        compression: Optional[BloscCompression],
+        attributes: Optional[Attributes],
+    ) -> zarr.Array:
+        compressor: Union[numcodecs.Blosc, Codec, None] = None
+        if compression is not None:
+            if group.metadata.zarr_format == 2:
+                compressor = compression.to_numcodecs()
+            else:
+                compressor = BloscCodec(**cast(Dict[str, Any], compression.to_config()))
+        return group.create_array(
+            name,
+            data=data,
+            chunks=chunks,
+            compressors=compressor,
+            attributes=None if attributes is None else attributes_dict(attributes),
+            # (store chunks equal to the fill value too, as zarr-python 2: identical records of both versions)
+            config={"write_empty_chunks": True},
+        )
+
+    def read(self, array: zarr.Array, selection: Selection) -> object:
+        if _direct_reads_enabled() and _is_basic_selection(selection) and (plan := _direct_read(array)) is not None:
+            return _read_direct(plan, array, selection)
+        return run(array.async_array.getitem(selection))
+
+    async def read_async(self, array: zarr.Array, selection: Selection) -> object:
+        if _direct_reads_enabled() and _is_basic_selection(selection) and (plan := _direct_read(array)) is not None:
+            return await _read_direct_async(plan, array, selection)
+        if _on_own_loop():
+            return await array.async_array.getitem(selection)
+        # (zarr-python's codec pipeline decodes on the calling event loop: run it on a worker thread's own loop)
+        return await asyncio.get_running_loop().run_in_executor(_read_executor(), self.read, array, selection)
+
+    def compression(self, array: zarr.Array) -> Optional[BloscCompression]:
+        if not (compressors := array.compressors):
+            return None
+        if len(compressors) == 1:
+            if isinstance(compressor := compressors[0], numcodecs.Blosc):  # (zarr format v2)
+                return BloscCompression.from_numcodecs(compressor)
+            if isinstance(compressor, BloscCodec):  # (zarr format v3)
+                return BloscCompression.from_config(cast(Dict[str, object], compressor.to_dict()["configuration"]))
+        raise ValueError(f"Unsupported compressors {compressors}")
+
+    # -- stores ----------------------------------------------------------------
+
+    def open_directory_store(self, path: str) -> ZarrStore:
+        return cast(ZarrStore, LocalStore(path))
+
+    def create_root_group(self, store: StoreLike, zarr_format: Literal[2, 3], attributes: Attributes) -> zarr.Group:
+        node = run(
+            zarr_async.create_group(
+                store=_zarr_store(store), zarr_format=zarr_format, attributes=attributes_dict(attributes)
+            )
+        )
+        return zarr.Group(node)
+
+    def open_store(self, store: StoreLike, consolidated: bool) -> zarr.Group:
+        zarr_store = _zarr_store(store)
+
+        if consolidated:
+            # the compressed consolidated metadata of zarr format v2 / v3 hierarchies (root zarr.json documents of zarr
+            # format v3 hierarchies with consolidated metadata are only stored compressed)
+            if (compressed := _get(zarr_store, V3_CONSOLIDATED_METADATA_KEY)) is not None:
+                return self._open_v3(zarr_store, decode_cbor_zlib(compressed), consolidated=True)
+            if (compressed := _get(zarr_store, V2_CONSOLIDATED_METADATA_KEY)) is None:
+                raise KeyError(V2_CONSOLIDATED_METADATA_KEY)
+        else:
+            if (compressed := _get(zarr_store, V3_CONSOLIDATED_METADATA_KEY)) is not None:
+                return self._open_v3(zarr_store, decode_cbor_zlib(compressed), consolidated=False)
+            if (root := _get(zarr_store, "zarr.json")) is not None:  # (zarr format v3 without consolidated metadata)
+                return self._open_v3(zarr_store, json.loads(root), consolidated=False)
+            return zarr.Group(
+                run(zarr_async.open_group(store=zarr_store, zarr_format=2, mode="r", use_consolidated=False))
+            )
+
+        document = decode_cbor_xz(compressed)
+        if (consolidated_format := document.get("zarr_consolidated_format")) != 1:
+            raise ValueError(f"Unsupported zarr consolidated metadata format: {consolidated_format}")
+        metadata = document["metadata"]
+
+        # the root group of the hierarchy, with the node metadata of all members (no node metadata reads)
+        prototype = default_buffer_prototype()
+        root = AsyncGroup._from_bytes_v2(
+            StorePath(zarr_store, ""),
+            prototype.buffer.from_bytes(json.dumps(metadata[".zgroup"]).encode()),
+            prototype.buffer.from_bytes(json.dumps(metadata.get(".zattrs", {})).encode()),
+            prototype.buffer.from_bytes(
+                json.dumps({**document, "metadata": _parent_grouped(metadata, _v2_node_path)}).encode()
+            ),
+        )
+        return zarr.Group(root)
+
+    def _open_v3(self, store: Store, document: Dict[str, Any], consolidated: bool) -> zarr.Group:
+        """Opens the root group of a zarr format v3 hierarchy from its root document"""
+        if consolidated:
+            nodes = document["consolidated_metadata"]
+            nodes = {**nodes, "metadata": _parent_grouped(nodes["metadata"], str)}
+            document = {**document, "consolidated_metadata": nodes}
+        else:
+            # (members are opened from their node metadata records)
+            document = {key: value for key, value in document.items() if key != "consolidated_metadata"}
+        with _ignored_warnings():
+            return zarr.Group(AsyncGroup.from_dict(StorePath(store, ""), document))
+
+    def consolidate_store(self, store: StoreLike) -> None:
+        zarr_store = _zarr_store(store)
+        if (root := _get(zarr_store, "zarr.json")) is not None:
+            # zarr format v3: the root document with the consolidated metadata of the hierarchy (as written by
+            # zarr-python's consolidate_metadata)
+            document = {
+                **json.loads(root),
+                "consolidated_metadata": {
+                    "kind": "inline",
+                    "must_understand": False,
+                    "metadata": _v3_node_documents(zarr_store),
+                },
+            }
+            value = encode_cbor_zlib(document)
+            run(zarr_store.set(V3_CONSOLIDATED_METADATA_KEY, default_buffer_prototype().buffer.from_bytes(value)))
+            return
+
+        metadata = {}
+        for key in _keys(zarr_store):
+            if is_v2_metadata_key(key) and (value := _get(zarr_store, key)) is not None:
+                metadata[key] = json.loads(value)
+        value = encode_cbor_xz({"zarr_consolidated_format": 1, "metadata": metadata})
+        run(zarr_store.set(V2_CONSOLIDATED_METADATA_KEY, default_buffer_prototype().buffer.from_bytes(value)))
+
+    def group_store(self, group: zarr.Group) -> StoreLike:
+        store = group.store
+        return store.store if isinstance(store, _TarStore) else cast(ZarrStore, store)
+
+
+backend = Backend()
diff --git a/ncore/impl/data/encoding.py b/ncore/impl/data/encoding.py
new file mode 100644
index 00000000..bc6b4ce8
--- /dev/null
+++ b/ncore/impl/data/encoding.py
@@ -0,0 +1,58 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Compressed binary encodings of JSON-like documents (record indices of indexed tar archives, compressed consolidated
+zarr metadata)"""
+
+from __future__ import annotations
+
+import io
+import lzma
+import zlib
+
+from typing import Any
+
+import cbor2
+
+
+def encode_cbor_xz(document: Any) -> bytes:
+    """Encodes a document as CBOR + LZMA (xz container format)"""
+    with io.BytesIO() as buffer:
+        with lzma.open(buffer, "wb", format=lzma.FORMAT_XZ) as lzma_file:
+            cbor2.dump(document, lzma_file)
+        return buffer.getvalue()
+
+
+def decode_cbor_xz(data: bytes) -> Any:
+    """Decodes a CBOR + LZMA (xz container format) encoded document"""
+    return cbor2.loads(lzma.LZMADecompressor().decompress(data))
+
+
+def encode_cbor_zlib(document: Any) -> bytes:
+    """Encodes a document as CBOR + zlib (decodes several times faster than CBOR + LZMA)"""
+    return zlib.compress(cbor2.dumps(document), 9)
+
+
+def decode_cbor_zlib(data: bytes) -> Any:
+    """Decodes a CBOR + zlib encoded document"""
+    return cbor2.loads(zlib.decompress(data))
+
+
+__all__ = [
+    "decode_cbor_xz",
+    "decode_cbor_zlib",
+    "encode_cbor_xz",
+    "encode_cbor_zlib",
+]
diff --git a/ncore/impl/data/itar.py b/ncore/impl/data/itar.py
new file mode 100644
index 00000000..855492da
--- /dev/null
+++ b/ncore/impl/data/itar.py
@@ -0,0 +1,404 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Indexed tar archives (``.itar``), independent of zarr-python.
+
+An indexed tar archive is a regular tar archive with a compressed record index appended after the tar end-of-archive
+blocks, enabling O(1) lookups of record payloads. :class:`IndexedTarStore` provides raw access to the records of such
+archives; zarr-python version specific store adapters (see :mod:`ncore.impl.data.stores`) store zarr hierarchies in
+them.
+
+Records are *write-once*: archives are append-only, and writing a record twice raises a ``ValueError``. The only
+exception are zarr node metadata records (see :func:`is_node_metadata_key`), which zarr-python 3 rewrites on every
+update of a node: they are kept in memory while writing and are appended (top-down) when closing the archive.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import io
+import logging
+import os
+import struct
+import tarfile
+import threading
+import weakref
+
+from enum import IntEnum, unique
+from pathlib import Path
+from typing import IO, Any, Callable, ClassVar, Dict, Iterator, List, Literal, NamedTuple, Optional, Tuple, Union, cast
+
+from upath import UPath
+
+from ncore.impl.data.encoding import decode_cbor_xz, encode_cbor_xz
+
+
+_logger = logging.getLogger(__name__)
+
+
+#: Names of the records of zarr node metadata (zarr format v2 and v3)
+NODE_METADATA_NAMES = (".zarray", ".zgroup", ".zattrs", "zarr.json")
+
+
+def is_node_metadata_key(key: str) -> bool:
+    """True for keys of zarr node metadata records (zarr format v2 and v3)"""
+    return key.rsplit("/", 1)[-1] in NODE_METADATA_NAMES
+
+
+class IndexedTarStore:
+    """The records of an indexed tar archive, opened for reading or (truncating) writing.
+
+    Parameters
+    ----------
+    itar_path : string, Path or UPath
+        Location of the archive (local or remote).
+    mode : string, optional, default 'r'
+        'r' to read an existing archive, or 'w' to create (truncate) and write an archive.
+    index_tail_read_size : int or None, optional, default 1 << 20 (1 MiB)
+        Maximum bytes read from the end of the archive in a single I/O call to load the record index. Covers both the
+        small index header and (ideally) the compressed index. Values smaller than the tar block size (512 bytes), or
+        None, result in separate reads of the index header and index, which is more robust for large indices.
+
+    Records are accessed via a mapping interface (``store[key]``, ``store[key] = value``, ``key in store``,
+    ``keys()``, ``len(store)``). Records of the index tail read are served from memory, so records stored near the end
+    of the archive don't repeat its high-latency I/O. Local archives are read via positional reads (``os.pread``),
+    which don't serialize reads of multiple threads.
+
+    Archives opened for writing need to be closed (``close()``, or by using the store as context manager), which
+    appends the node metadata records and the record index. Until then, node metadata records can be rewritten (see
+    the module documentation), and are served from memory.
+
+    All methods are thread-safe.
+    """
+
+    class Record(NamedTuple):
+        """The payload of a record within the archive file"""
+
+        offset: int
+        size: int
+
+    class TailBuffer(NamedTuple):
+        """Bytes read from the end of the archive file when loading the index"""
+
+        start: int  #: File offset of the first byte
+        data: bytes
+
+    @unique
+    class IndexType(IntEnum):
+        """Serialization formats of the record index"""
+
+        CBOR_LZMA_XZ_V1 = 1
+
+    #: Index header (little-endian): magic (4 bytes), index type (uint32), index offset (uint64), index size (uint32)
+    INDEX_HEADER_FORMAT: ClassVar[str] = "<4sIQI"
+    INDEX_HEADER_MAGIC: ClassVar[bytes] = b"itar"
+
+    #: Index type of written archives
+    INDEX_TYPE: ClassVar[IndexType] = IndexType.CBOR_LZMA_XZ_V1
+
+    #: Decoders / encoders of the supported index types (the index is a structure of arrays: record keys, payload
+    #: offsets and payload sizes, ordered by payload offsets)
+    INDEX_CODECS: ClassVar[Dict[IndexType, Tuple[Callable[[bytes], Any], Callable[[Any], bytes]]]] = {
+        IndexType.CBOR_LZMA_XZ_V1: (decode_cbor_xz, encode_cbor_xz),
+    }
+
+    def __init__(
+        self,
+        itar_path: Union[str, Path, UPath],
+        mode: Literal["r", "w"] = "r",
+        index_tail_read_size: Optional[int] = 1 << 20,  # 1 MiB by default
+    ) -> None:
+        if mode not in ("r", "w"):
+            raise ValueError(f"IndexedTarStore: unsupported mode {mode!r}, supported are 'r' and 'w'")
+        self._mode: Literal["r", "w"] = mode
+
+        # convert str / Path to an absolute UPath (with the UPath-internal `file://` protocol for local files)
+        itar_upath = UPath(itar_path)
+        if itar_upath.protocol == "":
+            itar_upath = UPath("file://" + str(itar_upath))
+        self._itar_upath = itar_upath.absolute()
+
+        # the tarfile module isn't thread-safe: writing and file-object based reading are serialized
+        self._lock = threading.RLock()
+
+        self._records: Dict[str, IndexedTarStore.Record] = {}
+        self._node_metadata: Dict[str, bytes] = {}  # (write mode, written when closing)
+        self._tail: Optional[IndexedTarStore.TailBuffer] = None
+        self._tar_file: Optional[tarfile.TarFile] = None
+        self._file: IO[bytes] = self._open_file(truncate=True)
+        if mode == "r":
+            self._records, self._tail = self._load_index(self._file, index_tail_read_size)
+        else:
+            self._tar_file = tarfile.TarFile(fileobj=self._file, mode="w")
+
+        self._pread_fd: Optional[int] = None
+        self._pread_finalizer: Optional[weakref.finalize] = None
+        self._open_positional_reads()
+
+    @property
+    def mode(self) -> Literal["r", "w"]:
+        """Access mode of the archive ('r' for reading, 'w' for writing)"""
+        return self._mode
+
+    @property
+    def itar_upath(self) -> UPath:
+        """Absolute location of the archive"""
+        return self._itar_upath
+
+    def __eq__(self, value: object) -> bool:
+        return isinstance(value, IndexedTarStore) and (self._itar_upath, self._mode) == (value._itar_upath, value._mode)
+
+    def __hash__(self) -> int:
+        return hash((self._itar_upath, self._mode))
+
+    def __repr__(self) -> str:
+        return f"IndexedTarStore('{self._itar_upath}', mode='{self._mode}')"
+
+    # -------------------------------------------------------------------------
+    # Records
+    # -------------------------------------------------------------------------
+
+    def keys(self) -> List[str]:
+        """Returns the keys of all records (in archive order, followed by node metadata records that are written when
+        closing the archive)"""
+        with self._lock:
+            return list(self._records) + list(self._node_metadata)
+
+    def __iter__(self) -> Iterator[str]:
+        return iter(self.keys())
+
+    def __len__(self) -> int:
+        with self._lock:
+            return len(self._records) + len(self._node_metadata)
+
+    def __contains__(self, key: object) -> bool:
+        with self._lock:
+            return key in self._records or key in self._node_metadata
+
+    def __getitem__(self, key: str) -> bytes:
+        """Reads the payload of a record (raises KeyError if it doesn't exist)"""
+        if self._node_metadata:
+            with self._lock:
+                if (value := self._node_metadata.get(key)) is not None:
+                    return value
+        record = self._records[key]
+
+        # (records end before the index, which ends the tail read: records starting within the tail read are part of it)
+        if (tail := self._tail) is not None and record.offset >= tail.start:
+            start = record.offset - tail.start
+            return tail.data[start : start + record.size]
+
+        if self._pread_fd is not None:
+            return self._pread(self._pread_fd, record.offset, record.size)
+
+        with self._lock:
+            # restore the position afterwards (the position to append records to when writing)
+            position = self._file.tell()
+            self._file.seek(record.offset)
+            value = self._file.read(record.size)
+            self._file.seek(position)
+            return value
+
+    @property
+    def is_remote(self) -> bool:
+        """True for archives opened for reading on asynchronous fsspec filesystems (e.g., s3fs, gcsfs, http), whose
+        records are read via asynchronous range requests by :meth:`read_async`"""
+        return self._async_fs() is not None
+
+    async def read_async(self, key: str) -> bytes:
+        """Reads the payload of a record asynchronously (raises KeyError if it doesn't exist).
+
+        Records of remote archives (see :attr:`is_remote`) are read via asynchronous range requests (concurrent reads
+        overlap), all other records synchronously (see :meth:`__getitem__`).
+        """
+        if (fs := self._async_fs()) is None or key in self._node_metadata:
+            return self[key]
+        record = self._records[key]
+        if ((tail := self._tail) is not None and record.offset >= tail.start) or record.size == 0:
+            return self[key]
+
+        request = fs._cat_file(self._itar_upath.path, start=record.offset, end=record.offset + record.size)
+        if (fs_loop := getattr(fs, "loop", None)) is None or fs_loop is asyncio.get_running_loop():
+            return await request
+        # (filesystems are bound to fsspec's I/O event loop, their sessions can't be used on other event loops)
+        return await asyncio.wrap_future(asyncio.run_coroutine_threadsafe(request, fs_loop))
+
+    def _async_fs(self) -> Any:
+        """The asynchronous fsspec filesystem of remote archives opened for reading (None otherwise)"""
+        if self._mode != "r" or self._pread_fd is not None:
+            return None
+        fs = self._itar_upath.fs
+        return fs if getattr(fs, "async_impl", False) and hasattr(fs, "_cat_file") else None
+
+    def __setitem__(self, key: str, value: Union[bytes, bytearray, memoryview]) -> None:
+        """Appends a record (requires write mode, raises ValueError if the record exists already). Node metadata
+        records are appended when closing the archive (and can be rewritten until then)."""
+        if self._mode != "w":
+            raise ValueError("IndexedTarStore: archive is opened for reading and doesn't support writing")
+        payload = bytes(value)
+
+        with self._lock:
+            if self._file.closed:
+                raise ValueError("IndexedTarStore: archive is closed")
+            if is_node_metadata_key(key):
+                self._node_metadata[key] = payload
+                return
+            if key in self._records:
+                raise ValueError(f"IndexedTarStore: record {key} already exists, records are write-once")
+            self._append(key, payload)
+
+    def _append(self, key: str, payload: bytes) -> None:
+        with self._lock:
+            # the header of a record can span multiple tar blocks (long keys), the payload is padded to full blocks
+            info = tarfile.TarInfo(key)
+            info.size = len(payload)
+            cast(tarfile.TarFile, self._tar_file).addfile(info, fileobj=io.BytesIO(payload))
+            padded_size = -(-len(payload) // tarfile.BLOCKSIZE) * tarfile.BLOCKSIZE
+            self._records[key] = self.Record(self._file.tell() - padded_size, len(payload))
+
+    def __delitem__(self, key: str) -> None:
+        raise NotImplementedError("IndexedTarStore: records can't be deleted")
+
+    # -------------------------------------------------------------------------
+    # Lifecycle
+    # -------------------------------------------------------------------------
+
+    def __enter__(self) -> IndexedTarStore:
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback) -> None:
+        self.close()
+
+    def close(self) -> None:
+        """Closes the archive (when writing: appends the record index, which is required to read the archive)"""
+        with self._lock:
+            if self._file.closed:
+                return
+            if self._mode == "w":
+                # (deterministic order independent of the order of node updates: parents before children)
+                node_metadata, self._node_metadata = self._node_metadata, {}
+                for key in sorted(node_metadata, key=lambda key: (key.count("/"), key)):
+                    self._append(key, node_metadata[key])
+                # closing the tar file appends the end-of-archive blocks, but keeps the file object open
+                cast(tarfile.TarFile, self._tar_file).close()
+                self._save_index(self._file, self._records)
+            self._file.close()
+            self._close_positional_reads()
+
+    def reload_resources(self) -> None:
+        """Re-opens the file objects (required in processes forked after opening the archive)"""
+        with self._lock:
+            position = self._file.tell()
+            self._file.close()
+            self._file = self._open_file(truncate=False)
+            self._file.seek(position)
+            if self._tar_file is not None:
+                self._tar_file.fileobj = self._file
+            self._close_positional_reads()
+            self._open_positional_reads()
+
+    def _open_file(self, truncate: bool) -> IO[bytes]:
+        if self._mode == "r":
+            return self._itar_upath.open("rb")
+        # writing requires a readable file (universal_pathlib <= 0.2.6, the last version supporting python 3.8,
+        # supports "wb+" / "rb+" without declaring them in its type annotations)
+        return self._itar_upath.open("wb+" if truncate else "rb+")  # type: ignore[call-overload]
+
+    def _open_positional_reads(self) -> None:
+        """Opens a file descriptor for positional reads of local archives opened for reading, which is closed with the
+        archive (or when the store is garbage-collected)"""
+        if self._mode == "r" and self._itar_upath.protocol in ("file", "local") and hasattr(os, "pread"):
+            self._pread_fd = os.open(self._itar_upath.path, os.O_RDONLY)
+            self._pread_finalizer = weakref.finalize(self, os.close, self._pread_fd)
+
+    def _close_positional_reads(self) -> None:
+        if self._pread_finalizer is not None:
+            self._pread_finalizer()
+        self._pread_finalizer = None
+        self._pread_fd = None
+
+    # -------------------------------------------------------------------------
+    # Positional reads / index serialization
+    # -------------------------------------------------------------------------
+
+    @staticmethod
+    def _pread(fd: int, offset: int, size: int) -> bytes:
+        value = os.pread(fd, size, offset)
+        while len(value) < size:  # positional reads may return fewer bytes than requested
+            if not (chunk := os.pread(fd, size - len(value), offset + len(value))):
+                raise EOFError(f"IndexedTarStore: archive ends within record at offset {offset}")
+            value += chunk
+        return value
+
+    @classmethod
+    def _load_index(
+        cls, file: IO[bytes], tail_read_size: Optional[int]
+    ) -> Tuple[Dict[str, IndexedTarStore.Record], IndexedTarStore.TailBuffer]:
+        """Loads the record index from the end of an archive file (with a single read if the index lies within the tail
+        read), and returns it together with the tail read"""
+        file.seek(0, os.SEEK_END)
+        file_size = file.tell()
+
+        tail_size = min(file_size, max(tarfile.BLOCKSIZE, tail_read_size or 0))
+        file.seek(file_size - tail_size)
+        tail = cls.TailBuffer(file_size - tail_size, file.read(tail_size))
+
+        # the index header is stored in the last tar block
+        header_offset = len(tail.data) - tarfile.BLOCKSIZE
+        magic, index_type, index_offset, index_size = struct.unpack_from(
+            cls.INDEX_HEADER_FORMAT, tail.data, header_offset
+        )
+        if magic != cls.INDEX_HEADER_MAGIC:
+            raise ValueError("IndexedTarStore: invalid index header, can't load indexed tar archive")
+        if index_type not in cls.INDEX_CODECS:
+            raise TypeError(f"IndexedTarStore: unsupported index type {index_type}")
+        decode_index = cls.INDEX_CODECS[cls.IndexType(index_type)][0]
+
+        if index_offset >= tail.start:
+            index = tail.data[index_offset - tail.start : index_offset - tail.start + index_size]
+        else:
+            file.seek(index_offset)
+            index = file.read(index_size)
+
+        table = decode_index(index)
+        _logger.debug(f"IndexedTarStore: loaded index of {len(table['items'])} records ({index_size} bytes)")
+        records = {
+            key: cls.Record(offset, size)
+            for key, offset, size in zip(table["items"], table["offset_datas"], table["sizes"])
+        }
+        return records, tail
+
+    @classmethod
+    def _save_index(cls, file: IO[bytes], records: Dict[str, IndexedTarStore.Record]) -> None:
+        """Appends the record index and its header to a finalized archive file (after the end-of-archive blocks)"""
+
+        def pad_to_block() -> None:
+            if remainder := file.tell() % tarfile.BLOCKSIZE:
+                file.write(tarfile.NUL * (tarfile.BLOCKSIZE - remainder))
+
+        index_offset = file.tell()
+        assert index_offset % tarfile.BLOCKSIZE == 0, "IndexedTarStore: archive end not at a tar block boundary"
+
+        table = sorted(((key, record.offset, record.size) for key, record in records.items()), key=lambda item: item[1])
+        keys, offsets, sizes = (list(column) for column in zip(*table)) if table else ([], [], [])
+        index = cls.INDEX_CODECS[cls.INDEX_TYPE][1]({"items": keys, "offset_datas": offsets, "sizes": sizes})
+        file.write(index)
+        pad_to_block()
+
+        file.write(
+            struct.pack(cls.INDEX_HEADER_FORMAT, cls.INDEX_HEADER_MAGIC, cls.INDEX_TYPE, index_offset, len(index))
+        )
+        pad_to_block()
+        _logger.debug(f"IndexedTarStore: saved index of {len(records)} records ({len(index)} bytes)")
diff --git a/ncore/impl/data/nodes.py b/ncore/impl/data/nodes.py
new file mode 100644
index 00000000..af63fde9
--- /dev/null
+++ b/ncore/impl/data/nodes.py
@@ -0,0 +1,356 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Groups and arrays of NCore data stores.
+
+:class:`Group` and :class:`Array` wrap zarr-python groups and arrays with a restricted, typed API, implemented by a
+backend of the installed zarr-python version (see :mod:`ncore.impl.data._backend`). The wrapped zarr-python objects
+are accessible via the ``zarr`` properties (for APIs of the installed zarr-python version not exposed by NCore).
+"""
+
+from __future__ import annotations
+
+import collections
+import threading
+
+from typing import TYPE_CHECKING, Any, Dict, Iterator, List, Literal, Optional, Tuple, Type, TypeVar, Union, cast
+
+import numpy as np
+import zarr
+
+from zarr import Array as ZarrArray
+from zarr import Group as ZarrGroup
+
+from ncore.impl.data._backend import Attributes, Backend, BloscCompression, Selection, attributes_dict
+from ncore.impl.data._json import JsonLike
+
+
+if TYPE_CHECKING:
+    import numpy.typing as npt  # type: ignore[import-not-found]
+
+
+def _installed_backend() -> Backend:
+    """The backend of the installed zarr-python version"""
+    if int(zarr.__version__.split(".", 1)[0]) >= 3:
+        from ncore.impl.data._zarr3 import backend as zarr3_backend
+
+        return zarr3_backend
+
+    from ncore.impl.data._zarr2 import backend as zarr2_backend
+
+    return zarr2_backend
+
+
+_backend: Backend = _installed_backend()
+
+#: zarr formats supported by the installed zarr-python version
+SUPPORTED_ZARR_FORMATS: Tuple[int, ...] = _backend.supported_zarr_formats
+
+
+def check_zarr_format(zarr_format: int) -> Literal[2, 3]:
+    """Validates that a zarr format is supported by the installed zarr-python version (raises ValueError otherwise)"""
+    if zarr_format not in SUPPORTED_ZARR_FORMATS:
+        raise ValueError(
+            f"zarr format v{zarr_format} is not supported by zarr-python {zarr.__version__} "
+            f"(supported: {', '.join(f'v{f}' for f in SUPPORTED_ZARR_FORMATS)})"
+        )
+    return 2 if zarr_format == 2 else 3
+
+
+T = TypeVar("T")
+
+
+class NodeCache:
+    """Cache of the opened nodes of a hierarchy opened for reading, by node path.
+
+    Opening a node constructs its zarr-python object (decoding its metadata), which dominates repeated reads of small
+    arrays (e.g., of frames). Cached nodes reference their metadata only, never array data. ``max_entries=None`` caches
+    all opened nodes, ``max_entries=N`` the N most recently used nodes (evicted nodes are re-opened on access).
+
+    All methods are thread-safe.
+    """
+
+    def __init__(self, max_entries: Optional[int] = None) -> None:
+        if max_entries is not None and max_entries < 1:
+            raise ValueError(f"max_entries needs to be positive or None (unbounded), got {max_entries}")
+        self._max_entries = max_entries
+        self._nodes: collections.OrderedDict[str, Union[ZarrGroup, ZarrArray]] = collections.OrderedDict()
+        self._lock = threading.Lock()
+
+    @property
+    def max_entries(self) -> Optional[int]:
+        return self._max_entries
+
+    def __len__(self) -> int:
+        return len(self._nodes)
+
+    def keys(self) -> List[str]:
+        """The paths of the cached nodes, from least to most recently used"""
+        with self._lock:
+            return list(self._nodes)
+
+    def _open(self, group: ZarrGroup, path: str) -> Union[ZarrGroup, ZarrArray]:
+        key = _join(group.path, path)
+        with self._lock:
+            if (node := self._nodes.get(key)) is not None:
+                self._nodes.move_to_end(key)
+                return node
+
+        # (concurrent first accesses open nodes redundantly, which is cheaper than serializing all accesses)
+        node = _backend.open_member(group, path)
+
+        with self._lock:
+            self._nodes[key] = node
+            if self._max_entries is not None:
+                while len(self._nodes) > self._max_entries:
+                    self._nodes.popitem(last=False)
+        return node
+
+
+class _Node:
+    """Common functionality of groups and arrays"""
+
+    def __init__(self, node: Union[ZarrGroup, ZarrArray]) -> None:
+        self._node = node
+
+    @property
+    def path(self) -> str:
+        """The path of the node within its store ("" for root groups)"""
+        return self._node.path
+
+    @property
+    def zarr_format(self) -> int:
+        """The zarr format of the node"""
+        return _backend.zarr_format(self._node)
+
+    @property
+    def attrs(self) -> Dict[str, JsonLike]:
+        """A copy of the attributes of the node"""
+        return cast(Dict[str, JsonLike], dict(self._node.attrs.asdict()))
+
+    def attr_str(self, name: str) -> str:
+        """Returns a string attribute (raises KeyError if missing, TypeError for other types)"""
+        return _typed_attr(self, name, str)
+
+    def attr_int(self, name: str) -> int:
+        """Returns an integer attribute (raises KeyError if missing, TypeError for other types)"""
+        return _typed_attr(self, name, int)
+
+    def attr_dict(self, name: str) -> Dict[str, JsonLike]:
+        """Returns a dictionary attribute (raises KeyError if missing, TypeError for other types)"""
+        return _typed_attr(self, name, dict)
+
+    def attr_list(self, name: str) -> List[JsonLike]:
+        """Returns a list attribute (raises KeyError if missing, TypeError for other types)"""
+        return _typed_attr(self, name, list)
+
+    def update_attrs(self, attributes: Attributes) -> None:
+        """Updates (merges) attributes of the node"""
+        self._node.attrs.update(attributes_dict(attributes))
+
+    def __repr__(self) -> str:
+        return f"{type(self).__name__}({self.path!r}, zarr_format={self.zarr_format})"
+
+
+class Array(_Node):
+    """An array of an NCore data store"""
+
+    def __init__(self, array: ZarrArray) -> None:
+        super().__init__(array)
+        self._array = array
+
+    @property
+    def zarr(self) -> ZarrArray:
+        """The wrapped zarr-python array"""
+        return self._array
+
+    @property
+    def shape(self) -> Tuple[int, ...]:
+        return tuple(self._array.shape)
+
+    @property
+    def dtype(self) -> np.dtype:
+        return np.dtype(self._array.dtype)
+
+    @property
+    def compression(self) -> Optional[BloscCompression]:
+        """The blosc compression of the array (None if uncompressed, raises ValueError for other compressors)"""
+        return _backend.compression(self._array)
+
+    def read(self, selection: Selection = Ellipsis) -> npt.NDArray[Any]:
+        """Reads (a selection of) the data of the array (0-dimensional selections result in 0-d arrays)"""
+        return np.asarray(_backend.read(self._array, selection))
+
+    async def read_async(self, selection: Selection = Ellipsis) -> npt.NDArray[Any]:
+        """Reads (a selection of) the data of the array asynchronously, see :meth:`read`.
+
+        Concurrent reads (e.g., via ``asyncio.gather``) overlap their I/O (e.g., range requests of remote archives) and
+        decompression, the running event loop isn't blocked (blocking work runs on worker threads).
+        """
+        return np.asarray(await _backend.read_async(self._array, selection))
+
+    def read_bytes(self) -> bytes:
+        """Reads a binary blob stored by :meth:`Group.create_bytes_array`"""
+        return self._blob(_backend.read(self._array, ()))
+
+    async def read_bytes_async(self) -> bytes:
+        """Reads a binary blob asynchronously, see :meth:`read_bytes` and :meth:`read_async`"""
+        return self._blob(await _backend.read_async(self._array, ()))
+
+    def _blob(self, value: object) -> bytes:
+        if isinstance(value, bytes):  # fixed-length bytes scalar (trailing NUL bytes are stripped)
+            return bytes(value)
+        if isinstance(value, np.ndarray) and value.dtype.kind == "S" and value.ndim == 0:
+            # (0-d fixed-length bytes arrays of some zarr-python 3 versions, zarr-python#4466)
+            return value.tobytes().rstrip(b"\x00")
+        if isinstance(value, np.ndarray) and value.dtype == np.uint8:
+            return value.tobytes()
+        raise TypeError(f"Array {self.path!r} doesn't store a binary blob ({self.dtype})")
+
+
+class Group(_Node):
+    """A group of an NCore data store.
+
+    Members of groups using a :class:`NodeCache` are opened via the cache (also members of member groups).
+    """
+
+    def __init__(self, group: ZarrGroup, cache: Optional[NodeCache] = None) -> None:
+        super().__init__(group)
+        self._group = group
+        self._cache = cache
+
+    @property
+    def zarr(self) -> ZarrGroup:
+        """The wrapped zarr-python group"""
+        return self._group
+
+    @property
+    def cache(self) -> Optional[NodeCache]:
+        """The node cache used to open members (None if not cached)"""
+        return self._cache
+
+    def with_cache(self, cache: Optional[NodeCache]) -> Group:
+        """Returns the group, using the given node cache to open its members"""
+        return Group(self._group, cache)
+
+    # -- reading ---------------------------------------------------------------
+
+    def _open(self, path: str) -> Union[ZarrGroup, ZarrArray]:
+        if self._cache is not None:
+            return self._cache._open(self._group, path)
+        return _backend.open_member(self._group, path)
+
+    def node(self, path: str) -> Union[Group, Array]:
+        """Returns a (nested) member group or array (raises KeyError if it doesn't exist)"""
+        node = self._open(path)
+        return Group(node, self._cache) if isinstance(node, ZarrGroup) else Array(node)
+
+    def group(self, path: str) -> Group:
+        """Returns a (nested) member group (raises KeyError if it doesn't exist or isn't a group)"""
+        if not isinstance(node := self._open(path), ZarrGroup):
+            raise KeyError(f"{_join(self.path, path)} is not a group")
+        return Group(node, self._cache)
+
+    def array(self, path: str) -> Array:
+        """Returns a (nested) member array (raises KeyError if it doesn't exist or isn't an array)"""
+        if not isinstance(node := self._open(path), ZarrArray):
+            raise KeyError(f"{_join(self.path, path)} is not an array")
+        return Array(node)
+
+    def members(self) -> List[str]:
+        """Names of all direct members"""
+        return _backend.member_names(self._group)
+
+    def __contains__(self, name: str) -> bool:
+        """True if the group has a direct member with the given name"""
+        return _backend.has_member(self._group, name)
+
+    def groups(self) -> Iterator[Tuple[str, Group]]:
+        """Iterates over the direct member groups (ordered by name)"""
+        for name, group in sorted(_backend.member_groups(self._group), key=lambda member: member[0]):
+            yield name, Group(group, self._cache)
+
+    # -- writing ---------------------------------------------------------------
+
+    def create_group(self, name: str, *, attributes: Optional[Attributes] = None) -> Group:
+        """Creates a member group (raises if a member with the name exists already)"""
+        return Group(_backend.create_group(self._group, name, attributes))
+
+    def require_group(self, name: str) -> Group:
+        """Returns a member group, creating it if it doesn't exist yet"""
+        return Group(_backend.require_group(self._group, name))
+
+    def create_array(
+        self,
+        name: str,
+        data: npt.ArrayLike,
+        *,
+        dtype: Optional[npt.DTypeLike] = None,
+        chunks: Union[Tuple[int, ...], Literal["auto"], None] = None,
+        compression: Optional[BloscCompression] = BloscCompression(),
+        attributes: Optional[Attributes] = None,
+    ) -> Array:
+        """Creates a member array initialized with data.
+
+        Parameters
+        ----------
+        name
+            Name of the array within the group.
+        data
+            Data of the array (defines the shape of the array, and its dtype if ``dtype`` isn't given).
+        dtype
+            Optional dtype of the array (the data is converted to it).
+        chunks
+            Chunk shape, a single chunk by default (zero-sized dimensions are clamped to 1). "auto" uses the chunk
+            shape heuristic of zarr-python.
+        compression
+            Blosc compression of the chunks (default: lz4, level 5, bit-shuffle), or None for uncompressed chunks.
+        attributes
+            Optional attributes of the array.
+        """
+        values = np.asarray(data, dtype=dtype)
+        if chunks != "auto":
+            chunks = tuple(max(1, int(size)) for size in (values.shape if chunks is None else chunks))
+        return Array(_backend.create_array(self._group, name, values, chunks, compression, attributes))
+
+    def create_bytes_array(self, name: str, data: bytes, *, attributes: Optional[Attributes] = None) -> Array:
+        """Stores a binary blob (e.g., an encoded image) as uncompressed member array.
+
+        zarr format v2: 0-dimensional fixed-length bytes array (``|S<N>``, the layout of ncore <= 19.8). zarr format v3:
+        1-dimensional ``uint8`` array (fixed-length bytes are not part of the zarr format v3 specification).
+        """
+        value = np.asarray(data) if self.zarr_format == 2 else np.frombuffer(data, dtype=np.uint8)
+        return self.create_array(name, value, compression=None, attributes=attributes)
+
+
+def _typed_attr(node: _Node, name: str, expected: Type[T]) -> T:
+    value = node._node.attrs[name]
+    if not isinstance(value, expected) or (isinstance(value, bool) and expected is int):
+        raise TypeError(f"Attribute {name!r} of {node.path!r} is not of type {expected.__name__}: {value!r}")
+    return cast(T, value)
+
+
+def _join(parent: str, path: str) -> str:
+    return f"{parent}/{path}" if parent else path
+
+
+__all__ = [
+    "Array",
+    "BloscCompression",
+    "Group",
+    "NodeCache",
+    "SUPPORTED_ZARR_FORMATS",
+    "check_zarr_format",
+]
diff --git a/ncore/impl/data/stores.py b/ncore/impl/data/stores.py
index a7a588f1..a48ffd4e 100644
--- a/ncore/impl/data/stores.py
+++ b/ncore/impl/data/stores.py
@@ -13,494 +13,70 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from __future__ import annotations
+"""Stores of NCore data: zarr hierarchies in indexed tar archives (:class:`IndexedTarStore`) or directories, with
+compressed consolidated metadata"""
 
-import io
-import logging
-import lzma
-import os
-import struct
-import tarfile
+from __future__ import annotations
 
-from dataclasses import dataclass, field
-from enum import IntEnum, auto, unique
 from pathlib import Path
-from threading import RLock
-from typing import IO, Any, Dict, Iterator, Literal, NamedTuple, Optional, Union, cast
+from typing import Union
 
-import cbor2
-import zarr
-
-from numcodecs import compat
 from upath import UPath
-from zarr._storage.store import Store
-from zarr.util import json_loads
-
-
-_logger = logging.getLogger(__name__)
-
-
-class IndexedTarStore(Store):
-    """A zarr store over *indexed* tar files
-
-    Parameters
-    ----------
-    itar_path : string
-        Location of the tar file (needs to end with '.itar').
-    mode : string, optional, default 'r'
-        One of 'r' to read an existing file, or 'w' to truncate and write a new
-        file.
-    index_tail_read_size : int or None, optional, default 1 << 20 (1 MiB)
-        Maximum bytes read from the tail of the file in a single I/O call to load the tar index.
-        Covers both the small index header and (ideally) the compressed index payload.
-        If not specified, perform separate reads for header and payload,
-        which is more robust for large indices, but requires two I/O calls.
-        Will be increased to at least one tar block size if specified smaller,
-        since the header is located in the last block.
-        Default 1 MiB size fits typical use cases, but may be increased if the compressed
-        index size is expected to be larger or decreased if remote access chunk sizes are smaller.
-
-    Read mode retains the byte range loaded for the index (the tail read) as an in-memory
-    cache. Any store key whose payload lies entirely in that range is served from the cache
-    without an additional read, so overlapping zarr chunks near the end of the archive do
-    not repeat the same high-latency tail I/O.
-
-    After modifying a IndexedTarStore, the ``close()`` method must be called, otherwise
-    essential data will not be written to the underlying files. The IndexedTarStore
-    class also supports the context manager protocol, which ensures the ``close()``
-    method is called on leaving the context, e.g.::
-
-        >>> with IndexedTarStore('data/array.itar', mode='w') as store:
-        ...     z = zarr.zeros((10, 10), chunks=(5, 5), store=store)
-        ...     z[...] = 42
-        ...     # no need to call store.close()
-
-    """
-
-    _erasable = False
-
-    @dataclass
-    class TarRecord:
-        """A file record within a tar file"""
-
-        offset_data: int
-        size: int
-
-    @dataclass
-    class TarRecordIndex:
-        """All file records within a tar file"""
-
-        records: Dict[str, IndexedTarStore.TarRecord] = field(default_factory=dict)
-
-    class TailBuffer(NamedTuple):
-        """Cached byte range from the tail of the archive file.
-
-        Retains the raw bytes of the single tail I/O read performed when the
-        index is loaded so that payloads falling entirely within that range can
-        be served from memory without an additional file read.
-        """
-
-        start: int  #: File byte offset where the tail read began
-        data: bytes  #: Raw bytes read from *start* to the end of the file
-
-    def __init__(
-        self,
-        itar_path: Union[str, Path, UPath],
-        mode: Literal["r", "w"] = "r",
-        index_tail_read_size: Optional[int] = 1 << 20,  # 1 MiB by default
-    ) -> None:
-        if mode not in ["r", "w"]:
-            raise ValueError("TarRecordIndex: only r/w modes supported")
-
-        # store properties
-        self.mode = mode
-
-        # Current understanding is that tarfile module in stdlib is not thread-safe,
-        # and so locking is required for both read and write. However, this has not
-        # been investigated in detail, perhaps no lock is needed if mode='r'.
-        self.mutex = RLock()
-
-        # convert str / Path to absolute UPath uncondtionally
-        itar_upath = UPath(itar_path)
-        if itar_upath.protocol == "":
-            # use UPath-internal `file://` protocol for local files
-            itar_upath = UPath("file://" + str(itar_upath))
-
-        self.itar_upath = itar_upath.absolute()
-
-        self._tail_buffer: Optional[IndexedTarStore.TailBuffer] = None
-
-        # open file object (require file to be both writeable and readable when writing) and
-        # tar file (only required for write mode, read mode directly uses file object)
-        self.tar_file_object: IO[Any]
-        self.tar_file: Optional[tarfile.TarFile] = None
-        if self.mode == "r":
-            self.tar_file_object = self.itar_upath.open("rb")
-        else:
-            # universal_path for Python 3.8 (<=0.2.6) doesn't expose a
-            # write/read mode in it's static type-hints, although "wb+" is still accepted
-            # if the FS supports it, so ignore type-checker here
-            self.tar_file_object = self.itar_upath.open("wb+")  # type: ignore[call-overload]
-
-            self.tar_file = tarfile.TarFile(fileobj=self.tar_file_object, mode="w")
-
-        # init / load index table
-        if self.mode == "r":
-            self.index, self._tail_buffer = self._load_tar_index(self.tar_file_object, index_tail_read_size)
-        else:
-            self.index = self.TarRecordIndex()
-
-    def __delitem__(self, _: str) -> None:
-        raise NotImplementedError("Deleting items is not supported")
-
-    def __iter__(self) -> Iterator[str]:
-        with self.mutex:
-            return iter(self.index.records.keys())
-
-    def __len__(self) -> int:
-        with self.mutex:
-            return len(self.index.records)
-
-    def __contains__(self, item: object) -> bool:
-        with self.mutex:
-            return item in self.index.records
-
-    def __getitem__(self, item: str) -> bytes:
-        with self.mutex:
-            # Query index for file record
-            record = self.index.records[item]  # raises KeyError if not in archive
-
-            # Check if payload is contained in tail buffer
-            if (tail := self._tail_buffer) is not None:
-                tail_start = tail.start
-                tail_end = tail_start + len(tail.data)
-                offset_data = record.offset_data
-                if offset_data >= tail_start and offset_data + record.size <= tail_end:
-                    return tail.data[offset_data - tail_start : offset_data - tail_start + record.size]
-
-            # Remember current tar file position
-            current_position = self.tar_file_object.tell()
-
-            # Read the value
-            self.tar_file_object.seek(record.offset_data)
-            value = self.tar_file_object.read(record.size)
-
-            # Return tar file to previous location
-            self.tar_file_object.seek(current_position)
-
-            return value
-
-    def __setitem__(self, item: str, value) -> None:
-        if self.mode != "w":
-            raise zarr.errors.ReadOnlyError
-
-        with self.mutex:
-            if item in self.index.records:
-                raise ValueError(f"{item} already exists, update is not supported")
-
-            value_bytes: bytes = compat.ensure_bytes(value)
-            value_size: int = len(value_bytes)
-
-            # Remember current tar file position, which is the start of the header
-            header_start_position = self.tar_file_object.tell()
-
-            # Store value in tar-file (will pre-pend a potentially *multi*-block header depending on item path-lengths)
-            tarinfo = tarfile.TarInfo(item)
-            tarinfo.size = value_size
-
-            cast(tarfile.TarFile, self.tar_file).addfile(tarinfo, fileobj=io.BytesIO(value_bytes))
-
-            # End position after writing both header and payload
-            end_position = self.tar_file_object.tell()
-
-            # Determine the effective value's payload size as a multiple of blocksize
-            payload_size = value_size
-            if remainder := payload_size % tarfile.BLOCKSIZE:
-                payload_size += tarfile.BLOCKSIZE - remainder
-
-            # Determine the effective header-size (can be multiple blocks for long path names)
-            header_size = end_position - header_start_position - payload_size
-
-            # Construct record from reconstructed size-information
-            record = self.TarRecord(
-                # Effective start of the data in the tar file (current tar file position + header-size)
-                header_start_position + header_size,
-                # Length of the data
-                value_size,
-            )
-
-            self.index.records[item] = record
-
-    def __enter__(self) -> IndexedTarStore:
-        return self
-
-    def __exit__(self, exc_type, exc_value, traceback) -> None:
-        self.close()
-
-    def close(self) -> None:
-        """Needs to be called after finishing updating the store"""
-        with self.mutex:
-            if self.mode == "w":
-                # Closing the tar file appends two finishing blocks to the end of the file
-                # if in write mode, but doesn't close the internal file object yet
-                cast(tarfile.TarFile, self.tar_file).close()
-
-                # Add index if writing
-                self._save_tar_index(self.tar_file_object, self.index)
-
-            self.tar_file_object.close()
-
-    def reload_resources(self) -> None:
-        """Reloads the tar file object *only* - useful to re-initialize the store in multi-process 'fork()' settings"""
-        with self.mutex:
-            # get current tar file path and seek positions, and close file object
-            current_position = self.tar_file_object.tell()
-            self.tar_file_object.close()
-
-            # reload file object (require file to be both writeable and readable when writing)
-            if self.mode == "r":
-                self.tar_file_object = self.itar_upath.open("rb")
-            else:
-                # universal_path for Python 3.8 (<=0.2.6) doesn't expose a
-                # write/read mode in it's static type-hints, although "rb+" is still accepted
-                # if the FS supports it, so ignore type-checker here
-                self.tar_file_object = self.itar_upath.open("rb+")  # type: ignore[call-overload]
-
-            if self.mode == "w":
-                # assign the new file object to the tar file (only required for write mode)
-                cast(tarfile.TarFile, self.tar_file).fileobj = self.tar_file_object
-
-            # seek to previous position
-            self.tar_file_object.seek(current_position)
-
-    # Methods / constants for storing index header and payload
-    INDEX_HEADER_MAGIC = b"itar"
-
-    # Index header binary format (20-bytes)
-    #
-    # <little-endian
-    # IndexMagic  - 4s - 4xchar             - 4bytes
-    # IndexType   - I  - unsigned int       - 4bytes
-    # IndexOffset - Q  - unsigned long long - 8bytes
-    # IndexSize   - I  - unsigned int       - 4bytes
-    INDEX_HEADER_FORMAT = "<4sIQI"
-
-    class IndexHeader(NamedTuple):
-        """A decoded index header"""
-
-        magic: bytes
-        type: int
-        offset: int
-        size: int
-
-    @unique
-    class IndexType(IntEnum):
-        """Enumerates different possible index storage types"""
-
-        CBOR_LZMA_XZ_V1 = auto()
-
-    @classmethod
-    def _load_tar_index(
-        cls, tar_file_object: IO[Any], index_tail_read_size: Optional[int]
-    ) -> tuple[TarRecordIndex, TailBuffer]:
-        """Loads a tar record index from the end of a tar file object.
-
-        Returns the parsed index and a :class:`TailBuffer` representing the
-        single tail read used to locate the index.  Callers may reuse the
-        buffer to satisfy reads for payloads that fall entirely within that range.
-        """
-        # Determine tail read size
-        if index_tail_read_size is None or index_tail_read_size < tarfile.BLOCKSIZE:
-            index_tail_read_size = tarfile.BLOCKSIZE  # explicit separate reads
-
-        original_file_position = tar_file_object.tell()
-
-        # Determine file size
-        tar_file_object.seek(0, os.SEEK_END)
-        file_size = tar_file_object.tell()
-
-        # Read up to index_tail_read_size from the tail in one read call
-        tail_buffer_size = min(file_size, index_tail_read_size)
-        tar_file_object.seek(file_size - tail_buffer_size)
-        tail_buffer = tar_file_object.read(tail_buffer_size)
-
-        # Extract the header from the last 512-byte block (it's guaranteed that the header fits into a single block)
-        header_binary_size = struct.calcsize(cls.INDEX_HEADER_FORMAT)
-        header_offset_in_tail_buffer = tail_buffer_size - tarfile.BLOCKSIZE
-        header = cls.IndexHeader._make(
-            struct.unpack(
-                cls.INDEX_HEADER_FORMAT,
-                tail_buffer[header_offset_in_tail_buffer : header_offset_in_tail_buffer + header_binary_size],
-            )
-        )
-
-        if header.magic != cls.INDEX_HEADER_MAGIC:
-            raise ValueError("IndexedTarStore: invalid index header, can't load indexed tar file")
-
-        # Try to extract the compressed index payload from the tail buffer
-        index_start_in_file = file_size - tail_buffer_size
-        if header.offset >= index_start_in_file:
-            # Payload is fully within the buffer we already read
-            index_start_in_tail_buffer = header.offset - index_start_in_file
-            index_binary = tail_buffer[index_start_in_tail_buffer : index_start_in_tail_buffer + header.size]
-        else:
-            # Rare: compressed index > index_tail_read_size — fall back to a second read
-            tar_file_object.seek(header.offset)
-            index_binary = tar_file_object.read(header.size)
-
-        tar_file_object.seek(original_file_position)
-
-        if header.type == cls.IndexType.CBOR_LZMA_XZ_V1.value:
-            _logger.debug(f"IndexedTarStore: lzma-compressed (xz archive format) index load size={len(index_binary)}")
-
-            # load table (SOA)
-            table = cbor2.loads(lzma.LZMADecompressor().decompress(index_binary))
-            items = table["items"]
-            offset_datas = table["offset_datas"]
-            sizes = table["sizes"]
-        else:
-            raise TypeError(f"IndexedTarStore: unsupported header type {header.type}")
-
-        # Construct record index from loaded table
-        index = cls.TarRecordIndex({item: cls.TarRecord(offset_datas[i], sizes[i]) for i, item in enumerate(items)})
-        return index, cls.TailBuffer(start=index_start_in_file, data=tail_buffer)
-
-    @classmethod
-    def _save_tar_index(cls, tar_file_object: IO[Any], index: TarRecordIndex) -> None:
-        """Saves a tar record index at the end of a tar file object (needs to be finalized / have two empty blocks appended already)"""
-
-        def fill_block() -> None:
-            # Fill up block with zeros
-            _, remainder = divmod(tar_file_object.tell(), tarfile.BLOCKSIZE)
-            if remainder > 0:
-                tar_file_object.write(tarfile.NUL * (tarfile.BLOCKSIZE - remainder))
-
-            assert tar_file_object.tell() % tarfile.BLOCKSIZE == 0, "Tar file not at block boundary"
-
-        # Remember where we are storing the index
-        index_offset = tar_file_object.tell()
-
-        assert index_offset % tarfile.BLOCKSIZE == 0, "Tar file not at block boundary"
-
-        # Reformat index table as SOA (sorted by offset)
-        table = [(item, record.offset_data, record.size) for (item, record) in index.records.items()]
-        items, offset_datas, sizes = list(zip(*sorted(table, key=lambda data: data[1]))) if len(table) else ([], [], [])
-
-        # Append compressed table to tar file
-        with io.BytesIO() as index_buffer:
-            # Compress table to in-memory buffer
-            with lzma.open(index_buffer, "wb", format=lzma.FORMAT_XZ) as lzma_file:
-                cbor2.dump({"items": items, "offset_datas": offset_datas, "sizes": sizes}, lzma_file)
-
-            index_binary = index_buffer.getvalue()
-            index_size = len(index_binary)
-
-            _logger.debug(f"IndexedTarStore lzma-compressed index store size={index_size}")
-
-            # Append buffer to tar file
-            tar_file_object.write(index_binary)
-
-            fill_block()
-
-        # Create index header block
-        assert struct.calcsize(cls.INDEX_HEADER_FORMAT) <= tarfile.BLOCKSIZE, (
-            "Index header larger than single block size"
-        )
-        header_binary = struct.pack(
-            cls.INDEX_HEADER_FORMAT,
-            cls.INDEX_HEADER_MAGIC,
-            cls.IndexType.CBOR_LZMA_XZ_V1.value,
-            index_offset,
-            index_size,
-        )
-        _logger.debug(f"IndexedTarStore: header store size={len(header_binary)}")
-
-        # Append index header to tar file
-        tar_file_object.write(header_binary)
-        fill_block()
-
-
-def consolidate_compressed_metadata(store: zarr.storage.BaseStore, metadata_key=".zmetadata.cbor.xz") -> None:
-    """Consolidate all metadata for groups and arrays within the given store
-    into a single compressed cbor resource and put it under the given key.
-
-    See Also
-    --------
-    zarr.consolidate_metadata
-    """
-    store = zarr.storage.normalize_store_arg(store, mode="w")
-
-    version = store._store_version
-
-    if version == 2:
-
-        def is_zarr_key(key) -> bool:
-            return key.endswith(".zarray") or key.endswith(".zgroup") or key.endswith(".zattrs")
-
-    else:
-        raise NotImplementedError("Only supporting V2 stores")
-
-    # Collect all meta-data
-    out = {
-        "zarr_consolidated_format": 1,
-        "metadata": {key: json_loads(store[key]) for key in store if is_zarr_key(key)},
-    }
 
-    with io.BytesIO() as metadata_buffer:
-        # Compress meta-data to in-memory buffer
-        with lzma.open(metadata_buffer, "wb") as lzma_file:
-            cbor2.dump(out, lzma_file)
+from ncore.impl.data._backend import (
+    V2_CONSOLIDATED_METADATA_KEY,
+    V3_CONSOLIDATED_METADATA_KEY,
+    Attributes,
+    StoreLike,
+    ZarrStore,
+)
+from ncore.impl.data.itar import IndexedTarStore
+from ncore.impl.data.nodes import Group, _backend, check_zarr_format
 
-        store[metadata_key] = metadata_buffer.getvalue()
 
+def open_directory_store(path: Union[str, Path, UPath]) -> ZarrStore:
+    """Opens a directory store (a directory per group / array, a file per record) for reading or writing"""
+    return _backend.open_directory_store(str(path))
 
-class ConsolidatedCompressedMetadataStore(zarr.storage.ConsolidatedMetadataStore):
-    """A layer over other storage, where the metadata has been consolidated into a single compressed key."""
 
-    # Overwrite constructor to perform decompression of metadata
-    def __init__(self, store: zarr.storage.StoreLike, metadata_key=".zmetadata.cbor.xz") -> None:
-        self.store = Store._ensure_store(store)
+def create_root_group(store: StoreLike, attributes: Attributes, *, zarr_format: int = 2) -> Group:
+    """Creates the root group of a hierarchy in a store, in zarr format v2 (default, readable with all zarr-python
+    versions) or v3 (requires zarr-python 3 for writing and reading)"""
+    return Group(_backend.create_root_group(store, check_zarr_format(zarr_format), attributes))
 
-        # retrieve consolidated metadata
-        meta = cbor2.loads(lzma.LZMADecompressor().decompress(self.store[metadata_key]))
 
-        # check format of consolidated metadata
-        consolidated_format = meta.get("zarr_consolidated_format", None)
-        if consolidated_format != 1:
-            raise zarr.MetadataError("unsupported zarr consolidated metadata format: %s" % consolidated_format)
+def open_store(store: StoreLike, consolidated: bool = True) -> Group:
+    """Opens the root group of the hierarchy of a store for reading, via its compressed consolidated metadata if
+    consolidated (raises KeyError if the store has no compressed consolidated metadata)"""
+    return Group(_backend.open_store(store, consolidated))
 
-        # decode metadata
-        self.meta_store: zarr.storage.Store = zarr.KVStore(meta["metadata"])
 
+def consolidate_store(store: StoreLike) -> None:
+    """Stores the compressed consolidated metadata of the hierarchy of a store (after writing all nodes)"""
+    _backend.consolidate_store(store)
 
-def open_compressed_consolidated(
-    store: zarr.storage.StoreLike, metadata_key=".zmetadata.cbor.xz", mode="r+", **kwargs
-) -> zarr.hierarchy.Group:
-    """Open group using metadata previously consolidated and compressed into a single key.
 
-    See Also
-    --------
-    consolidate_compressed_metadata
-    zarr.open_consolidated
-    """
+def close_store(store: StoreLike) -> None:
+    """Closes a store (required for indexed tar stores opened for writing)"""
+    if isinstance(store, IndexedTarStore):
+        store.close()
 
-    # normalize parameters
-    zarr_version = kwargs.get("zarr_version")
-    store = zarr.storage.normalize_store_arg(
-        store, storage_options=kwargs.get("storage_options"), mode=mode, zarr_version=zarr_version
-    )
-    if mode not in {"r", "r+"}:
-        raise ValueError("invalid mode, expected either 'r' or 'r+'; found {!r}".format(mode))
 
-    path = kwargs.pop("path", None)
-    if store._store_version == 2:
-        ConsolidatedStoreClass = ConsolidatedCompressedMetadataStore
-    else:
-        raise NotImplementedError("Only supporting V2 stores")
+def group_store(group: Group) -> StoreLike:
+    """The store of a group (the :class:`IndexedTarStore` of groups of indexed tar archives)"""
+    return _backend.group_store(group.zarr)
 
-    # setup metadata store
-    meta_store = ConsolidatedStoreClass(store, metadata_key=metadata_key)
 
-    # pass through
-    chunk_store = kwargs.pop("chunk_store", None) or store
-    return zarr.convenience.open(store=meta_store, chunk_store=chunk_store, mode=mode, path=path, **kwargs)
+__all__ = [
+    "IndexedTarStore",
+    "StoreLike",
+    "V2_CONSOLIDATED_METADATA_KEY",
+    "V3_CONSOLIDATED_METADATA_KEY",
+    "ZarrStore",
+    "close_store",
+    "consolidate_store",
+    "create_root_group",
+    "group_store",
+    "open_directory_store",
+    "open_store",
+]
diff --git a/ncore/impl/data/stores_test.py b/ncore/impl/data/stores_test.py
index 030cb0c6..f3029dc0 100644
--- a/ncore/impl/data/stores_test.py
+++ b/ncore/impl/data/stores_test.py
@@ -13,195 +13,1182 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import itertools
+import asyncio
+import concurrent.futures
+import gc
+import hashlib
+import json
+import os
+import struct
+import tarfile
 import tempfile
+import threading
 import unittest
+import zlib
 
 from pathlib import Path
-from typing import IO, Any, Optional
+from typing import Any, Dict, List, Optional, Tuple, cast
+from unittest import mock
 
+import cbor2
 import numpy as np
 import parameterized
 import zarr
 
-from ncore.impl.data.stores import IndexedTarStore, consolidate_compressed_metadata, open_compressed_consolidated
+from ncore.impl.data import nodes
+from ncore.impl.data.encoding import decode_cbor_xz
+from ncore.impl.data.itar import is_node_metadata_key
+from ncore.impl.data.stores import (
+    V2_CONSOLIDATED_METADATA_KEY,
+    V3_CONSOLIDATED_METADATA_KEY,
+    IndexedTarStore,
+    StoreLike,
+    close_store,
+    consolidate_store,
+    create_root_group,
+    group_store,
+    open_directory_store,
+    open_store,
+)
 
 
-COMPRESSED_CONSOLIDATED_VALUES = [False, True]
-INDEX_TAIL_READ_SIZES = [
-    None,
-    1 << 10,
-    1 << 20,
-    1,
-    512,
-    513,
-]  # None (explicit separate reads), 0.5 MiB, 1 MiB (default), 1 byte / 512 / 513 bytes (edge cases resulting in separate reads)
+#: None / values < 512 (separate reads of index header and index), 1 KiB, 1 MiB (default), edge cases around 512
+INDEX_TAIL_READ_SIZES = [None, 1, 512, 513, 1 << 10, 1 << 20]
+
+_ROOT_ATTRIBUTES = {"some": "thing"}
+
+ZARR_PYTHON_3 = int(zarr.__version__.split(".", 1)[0]) >= 3
+
+#: All zarr formats (tests of zarr formats unsupported by the installed zarr-python version are skipped)
+ZARR_FORMATS = (2, 3)
+
+
+def _require(test: unittest.TestCase, zarr_format: int) -> None:
+    if zarr_format not in nodes.SUPPORTED_ZARR_FORMATS:
+        test.skipTest(f"zarr format v{zarr_format} requires zarr-python 3")
+
+
+def _fill_reference(root: nodes.Group) -> None:
+    rng = np.random.default_rng(0)
+    root.create_array("foo", rng.random((3, 3, 3)), chunks=(2, 3, 3))
+    sub = root.create_group("subgroup", attributes={"other": [1, 2, 3]})
+    sub.create_array("foo", rng.random((5, 5, 5)), attributes={"unit": "m"})
+    sub.create_array("empty", np.zeros((0, 3), dtype=np.float32))
+    sub.create_bytes_array("blob", b"\x89PNG\x00some binary data", attributes={"format": "png"})
+
+
+def _check_reference(test: unittest.TestCase, root: nodes.Group) -> None:
+    rng = np.random.default_rng(0)
+    np.testing.assert_array_equal(root.array("foo").read(), rng.random((3, 3, 3)))
+    np.testing.assert_array_equal(root.array("subgroup/foo").read(), rng.random((5, 5, 5)))
+    test.assertEqual(root.array("subgroup/empty").shape, (0, 3))
+    test.assertEqual(root.array("subgroup/blob").read_bytes(), b"\x89PNG\x00some binary data")
+    test.assertEqual(root.attrs, _ROOT_ATTRIBUTES)
+    test.assertEqual(root.group("subgroup").attrs, {"other": [1, 2, 3]})
+    test.assertEqual(root.array("subgroup/foo").attrs, {"unit": "m"})
+    test.assertEqual(sorted(root.members()), ["foo", "subgroup"])
+    test.assertEqual(sorted(root.group("subgroup").members()), ["blob", "empty", "foo"])
+    test.assertEqual([name for name, _ in root.groups()], ["subgroup"])
+    test.assertIn("blob", root.group("subgroup"))
+    test.assertNotIn("missing", root.group("subgroup"))
+    with test.assertRaises(KeyError):
+        root.group("foo")
+    with test.assertRaises(KeyError):
+        root.array("subgroup")
+    with test.assertRaises(KeyError):
+        root.array("missing")
+
+
+def _decode_v3_root(value: bytes) -> Any:
+    return cbor2.loads(zlib.decompress(value))
+
+
+def _write(store: StoreLike, consolidate: bool = True, zarr_format: int = 2) -> None:
+    _fill_reference(create_root_group(store, _ROOT_ATTRIBUTES, zarr_format=zarr_format))
+    if consolidate:
+        consolidate_store(store)
+    close_store(store)
 
 
 class TestIndexedTarStore(unittest.TestCase):
-    """Test to verify functionality of IndexedTarStore"""
+    """Records of indexed tar archives"""
 
     def setUp(self):
-        # Fill a reference group with an in-memory store
-        self.g_ref = zarr.open(store=zarr.MemoryStore())
-        self.g_ref.create_dataset("foo", data=np.random.default_rng().random((3, 3, 3)))
-        self.g_ref.attrs.update({"some": "thing"})
-        self.g_ref.require_group("subgroup").create_dataset("foo", data=np.random.default_rng().random((5, 5, 5)))
-
-    def check_with_reference(self, group):
-        """Verifies all values of a group against the reference"""
-        self.assertIsNone(np.testing.assert_array_equal(self.g_ref["foo"][()], group["foo"][()]))
-        self.assertIsNone(
-            np.testing.assert_array_equal(self.g_ref["subgroup"]["foo"][()], group["subgroup"]["foo"][()])
-        )
-        self.assertDictEqual(self.g_ref.attrs.asdict(), group.attrs.asdict())
+        self._tmp = tempfile.TemporaryDirectory()
+        self.path = os.path.join(self._tmp.name, "store.itar")
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def test_records(self):
+        records = {"a": b"0123456789", "dir/b": b"x" * 513, "empty": b"", "l" * 150 + "/long-key": bytes(range(256))}
+        with IndexedTarStore(self.path, mode="w") as store:
+            for key, value in records.items():
+                store[key] = value
+            self.assertEqual(store["dir/b"], records["dir/b"])  # readable while writing
+            with self.assertRaises(ValueError):  # write-once
+                store["a"] = b"again"
+
+        store = IndexedTarStore(self.path)
+        self.assertEqual(store.keys(), list(records))
+        self.assertEqual(len(store), len(records))
+        self.assertEqual({key: store[key] for key in store}, records)
+        self.assertIn("a", store)
+        self.assertNotIn("missing", store)
+        with self.assertRaises(KeyError):
+            store["missing"]
+        with self.assertRaises(ValueError):  # read-only
+            store["c"] = b"c"
+        with self.assertRaises(NotImplementedError):
+            del store["a"]
+
+        # archives are regular tar archives
+        with tarfile.open(self.path) as tar:
+            self.assertEqual({m.name: _read_member(tar, m) for m in tar.getmembers()}, records)
+
+    @parameterized.parameterized.expand([(size,) for size in INDEX_TAIL_READ_SIZES])
+    def test_index_tail_read_sizes(self, index_tail_read_size: Optional[int]):
+        records = {f"r{i}": bytes([i]) * (i * 97) for i in range(40)}
+        with IndexedTarStore(self.path, mode="w") as store:
+            for key, value in records.items():
+                store[key] = value
+
+        store = IndexedTarStore(self.path, index_tail_read_size=index_tail_read_size)
+        self.assertEqual({key: store[key] for key in store}, records)
+
+    def test_records_within_the_index_tail_read_require_no_reads(self):
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["first"] = b"1" * 1000
+            store["last"] = b"2" * 1000
+
+        store = IndexedTarStore(self.path, index_tail_read_size=Path(self.path).stat().st_size)
+        with mock.patch("os.pread", side_effect=AssertionError("unexpected read")):
+            self.assertEqual(store["first"], b"1" * 1000)
+            self.assertEqual(store["last"], b"2" * 1000)
+
+    def test_records_partially_within_the_index_tail_read(self):
+        """Records starting before the index tail read are read from the archive file"""
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["large"] = bytes(range(256)) * 64  # 16 KiB
+            store["small"] = b"s"
+
+        file_size = Path(self.path).stat().st_size
+        store = IndexedTarStore(self.path, index_tail_read_size=file_size - 1024)  # ends within "large"
+        self.assertEqual(store["large"], bytes(range(256)) * 64)
+        self.assertEqual(store["small"], b"s")
+
+    def test_positional_reads(self):
+        """Local archives are read via positional reads"""
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["a"] = b"a" * 4096
+
+        store = IndexedTarStore(self.path, index_tail_read_size=None)
+        with mock.patch("os.pread", side_effect=os.pread) as pread:
+            self.assertEqual(store["a"], b"a" * 4096)
+        self.assertGreater(pread.call_count, 0)
+
+    def test_dropped_stores_release_file_descriptors(self):
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["a"] = b"a"
+
+        def open_fds() -> int:
+            return len(os.listdir(f"/proc/{os.getpid()}/fd"))
+
+        IndexedTarStore(self.path)["a"]
+        gc.collect()
+        before = open_fds()
+        for _ in range(20):
+            IndexedTarStore(self.path)["a"]
+        gc.collect()
+        self.assertEqual(open_fds(), before)
+
+    def test_reload_resources(self):
+        """Reloading resources neither truncates archives opened for writing nor affects reads"""
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["first"] = b"first payload"
+            store.reload_resources()
+            self.assertEqual(store["first"], b"first payload")
+            store["second"] = b"second payload"
+
+        store = IndexedTarStore(self.path, index_tail_read_size=None)
+        store.reload_resources()
+        self.assertEqual(store["first"], b"first payload")
+        self.assertEqual(store["second"], b"second payload")
+
+    def test_invalid_archives(self):
+        invalid_mode: Any = "a"
+        with self.assertRaises(ValueError):
+            IndexedTarStore(self.path, mode=invalid_mode)
+        Path(self.path).write_bytes(b"\0" * 2048)
+        with self.assertRaises(ValueError):
+            IndexedTarStore(self.path)
+
+    def test_unsupported_index_types(self):
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["a"] = b"a"
+        data = bytearray(Path(self.path).read_bytes())
+        header = len(data) - tarfile.BLOCKSIZE  # (the index header is stored in the last tar block)
+        magic, _, offset, size = struct.unpack_from(IndexedTarStore.INDEX_HEADER_FORMAT, data, header)
+        struct.pack_into(IndexedTarStore.INDEX_HEADER_FORMAT, data, header, magic, 99, offset, size)
+        Path(self.path).write_bytes(bytes(data))
+        with self.assertRaisesRegex(TypeError, "unsupported index type 99"):
+            IndexedTarStore(self.path)
+
+    def test_empty(self):
+        with IndexedTarStore(self.path, mode="w"):
+            pass
+        self.assertEqual(IndexedTarStore(self.path).keys(), [])
+
+    def test_node_metadata_records(self):
+        """Node metadata records can be rewritten while writing, they are appended when closing (parents first)"""
+        with IndexedTarStore(self.path, mode="w") as store:
+            store["g/a/zarr.json"] = b"a v1"
+            store["g/a/c/0"] = b"chunk"
+            store["zarr.json"] = b"root v1"
+            store["g/a/zarr.json"] = b"a v2"  # rewritten
+            store[".zattrs"] = b"attrs"
+            self.assertEqual(store["g/a/zarr.json"], b"a v2")  # readable while writing
+            self.assertEqual(len(store), 4)
+            self.assertIn("zarr.json", store)
+            with self.assertRaises(ValueError):  # other records are write-once
+                store["g/a/c/0"] = b"again"
+        with self.assertRaises(ValueError):  # closed
+            store["zarr.json"] = b"root v2"
+
+        store = IndexedTarStore(self.path)
+        self.assertEqual(store.keys(), ["g/a/c/0", ".zattrs", "zarr.json", "g/a/zarr.json"])
+        self.assertEqual(store["g/a/zarr.json"], b"a v2")
+        self.assertEqual(store["zarr.json"], b"root v1")
+        with tarfile.open(self.path) as tar:  # (written once each)
+            self.assertEqual([m.name for m in tar.getmembers()], store.keys())
+
+    def test_node_metadata_keys(self):
+        for key in (".zarray", ".zgroup", ".zattrs", "zarr.json", "a/b/.zarray", "a/zarr.json"):
+            self.assertTrue(is_node_metadata_key(key), key)
+        for key in ("zarr.json.cbor.zlib", ".zmetadata.cbor.xz", "a/0.0", "a/c/0/0", "a/x.zarray", "a/zarr.json/0"):
+            self.assertFalse(is_node_metadata_key(key), key)
+
+
+class TestStores(unittest.TestCase):
+    """zarr hierarchies in indexed tar archives and directories"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        self.itar_path = os.path.join(self._tmp.name, "store.itar")
+        self.directory_path = os.path.join(self._tmp.name, "store.zarr")
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def _store(self, store_type: str, mode: str) -> StoreLike:
+        if store_type == "itar":
+            return IndexedTarStore(self.itar_path, mode="w" if mode == "w" else "r")
+        return open_directory_store(self.directory_path)
 
     @parameterized.parameterized.expand(
-        itertools.product(
-            INDEX_TAIL_READ_SIZES,
-        )
+        [
+            (store_type, consolidated, zarr_format)
+            for store_type in ("itar", "directory")
+            for consolidated in (True, False)
+            for zarr_format in ZARR_FORMATS
+        ]
     )
-    def test_reserialization(self, index_tail_read_size: Optional[int]):
-        """Make sure storing / loading of regular zarr data to .itar files works correctly"""
+    def test_reserialization(self, store_type: str, consolidated: bool, zarr_format: int):
+        _require(self, zarr_format)
+        _write(self._store(store_type, "w"), zarr_format=zarr_format)
 
-        # re-serialize to .itar archive
-        with tempfile.NamedTemporaryFile(suffix=".itar") as f:
-            with IndexedTarStore(f.name, mode="w") as s_itar_out:  # closes file on exit
-                zarr.copy_store(self.g_ref.store, s_itar_out)
+        store = self._store(store_type, "r")
+        root = open_store(store, consolidated)
+        _check_reference(self, root)
+        self.assertEqual(root.zarr_format, zarr_format)
+        self.assertEqual(root.array("subgroup/foo").zarr_format, zarr_format)
 
-            # reload store from file
-            store = IndexedTarStore(f.name, index_tail_read_size=index_tail_read_size)
-            g_reload = zarr.open(store=store, mode="r")
+        if isinstance(store, IndexedTarStore):
+            self.assertIs(group_store(root), store)
+            self.assertIs(group_store(root.group("subgroup")), store)
+            store.reload_resources()
+            _check_reference(self, root)
 
-            # check all data was correctly serialized / deserialized
-            self.check_with_reference(g_reload)
+    @parameterized.parameterized.expand(
+        [(size, zarr_format) for size in INDEX_TAIL_READ_SIZES for zarr_format in ZARR_FORMATS]
+    )
+    def test_consolidated_open_reads_no_node_metadata(self, index_tail_read_size: Optional[int], zarr_format: int):
+        _require(self, zarr_format)
+        _write(self._store("itar", "w"), zarr_format=zarr_format)
 
-            # check reloading resources is functional
-            store.reload_resources()
-            self.check_with_reference(g_reload)
+        store = IndexedTarStore(self.itar_path, index_tail_read_size=index_tail_read_size)
+        with mock.patch.object(
+            IndexedTarStore, "__getitem__", autospec=True, side_effect=IndexedTarStore.__getitem__
+        ) as getitem:
+            _check_reference(self, open_store(store, consolidated=True))
+        read_keys = [call.args[1] for call in getitem.call_args_list]
+        self.assertIn(V2_CONSOLIDATED_METADATA_KEY if zarr_format == 2 else V3_CONSOLIDATED_METADATA_KEY, read_keys)
+        self.assertEqual([key for key in read_keys if is_node_metadata_key(key)], [])
 
     @parameterized.parameterized.expand(
-        itertools.product(
-            INDEX_TAIL_READ_SIZES,
-        )
+        [(store_type, zarr_format) for store_type in ("itar", "directory") for zarr_format in ZARR_FORMATS]
     )
-    def test_compressed_consolidated(self, index_tail_read_size: Optional[int]):
-        """Make sure compressed consolidated meta data is stored/loaded correctly"""
+    def test_missing_consolidated_metadata(self, store_type: str, zarr_format: int):
+        _require(self, zarr_format)
+        _write(self._store(store_type, "w"), consolidate=False, zarr_format=zarr_format)
+        with self.assertRaises(KeyError):
+            open_store(self._store(store_type, "r"), consolidated=True)
+        _check_reference(self, open_store(self._store(store_type, "r"), consolidated=False))
 
-        # serialize to .itar archive (will also serialize compressed-consolidated meta-data)
-        with tempfile.NamedTemporaryFile(suffix=".itar") as f:
-            with IndexedTarStore(f.name, mode="w") as s_itar_out:  # closes file on exit
-                zarr.copy_store(self.g_ref.store, s_itar_out)
+    @parameterized.parameterized.expand([(zarr_format,) for zarr_format in ZARR_FORMATS])
+    def test_unconsolidated_open_reads_node_metadata(self, zarr_format: int):
+        """Hierarchies opened without consolidated metadata are resolved from their node metadata records (e.g., nodes
+        added after consolidating, as component writers extending existing directory stores)"""
+        _require(self, zarr_format)
+        root = create_root_group(store := self._store("directory", "w"), _ROOT_ATTRIBUTES, zarr_format=zarr_format)
+        root.create_array("before", np.zeros(2))
+        consolidate_store(store)
+        root.create_array("after", np.ones(2))
 
-                # consolidate compress meta-data
-                consolidate_compressed_metadata(s_itar_out)
+        self.assertEqual(
+            sorted(open_store(self._store("directory", "r"), consolidated=False).members()), ["after", "before"]
+        )
+        self.assertEqual(sorted(open_store(self._store("directory", "r"), consolidated=True).members()), ["before"])
 
-            # reload store from file with compressed consolidated meta-data
-            store = IndexedTarStore(f.name, index_tail_read_size=index_tail_read_size)
-            g_reload = open_compressed_consolidated(store=store, mode="r")
+    @parameterized.parameterized.expand([(zarr_format,) for zarr_format in ZARR_FORMATS])
+    def test_stores_are_identical(self, zarr_format: int):
+        """Indexed tar archives and directories store the same records"""
+        _require(self, zarr_format)
+        _write(self._store("itar", "w"), zarr_format=zarr_format)
+        _write(self._store("directory", "w"), zarr_format=zarr_format)
 
-            # check all data was correctly serialized / deserialized
-            self.check_with_reference(g_reload)
+        archive = IndexedTarStore(self.itar_path)
+        files = {
+            str(path.relative_to(self.directory_path)): path.read_bytes()
+            for path in Path(self.directory_path).rglob("*")
+            if path.is_file()
+        }
+        self.assertEqual(sorted(archive.keys()), sorted(files))
+        for key in archive.keys():
+            if key == V2_CONSOLIDATED_METADATA_KEY:  # (ordered by the record order of the stores)
+                self.assertEqual(decode_cbor_xz(archive[key]), decode_cbor_xz(files[key]))
+            elif key == V3_CONSOLIDATED_METADATA_KEY:
+                self.assertEqual(_decode_v3_root(archive[key]), _decode_v3_root(files[key]))
+            else:
+                self.assertEqual(archive[key], files[key], key)
 
-            # check reloading resources is functional
-            store.reload_resources()
-            self.check_with_reference(g_reload)
+    def test_directory_stores_are_not_truncated(self):
+        Path(self.directory_path).mkdir()
+        (Path(self.directory_path) / "other").write_bytes(b"other")
+        _write(self._store("directory", "w"))
+        self.assertEqual((Path(self.directory_path) / "other").read_bytes(), b"other")
+
+    #: sha256 of the records of the hierarchy of :meth:`_write_golden` written by ncore 19.8 (zarr-python 2)
+    GOLDEN_RECORDS = {
+        ".zattrs": "0e326aaa71db8ecc9051e106a4a70695515ad95a95ff348445699fd79f907623",
+        ".zgroup": "2383746e67b4bcc2762b3f100f06c3fa2d5f149ab5a8e5da5d33521464a01959",
+        ".zmetadata.cbor.xz": "245a85421d308c67ea2e452746f0628fe0b3bb3a337af3fec3b46e75b044a4cc",
+        "foo/.zarray": "28db182434d2702c2ccc384e0c1d386359f308bd275821655377b09a40c610fa",
+        "foo/0.0.0": "e9361a616760d41ac367d54b2349f18184e8a226563ca34047e691cf393a9610",
+        "foo/1.0.0": "8611647425ac52bdfd586f7342bdb768d363ae139aadafa7130c7a4368015614",
+        "subgroup/.zattrs": "b638e55387a56bef81862ccd44889e0068a67b04cf00a4c4e90783cf2940e160",
+        "subgroup/.zgroup": "2383746e67b4bcc2762b3f100f06c3fa2d5f149ab5a8e5da5d33521464a01959",
+        "subgroup/bar/.zarray": "6c83cd89670edfe4f9f12cf48f7de5f55ed9cefbfa38b168f1ab14dfb082c449",
+        "subgroup/bar/0": "3c7acfa845b57df9e3a46779d4f17c7eb9d697d63dd8b2c30c176c6fec90051b",
+        "subgroup/blob/.zarray": "dcb9ebabd3bb2d803fd59453b16ba414f6d55d4451ba597170b4a90fae4e749b",
+        "subgroup/blob/0": "5399a4c8461dd3fd590854ff73ec4e88db9462165e9160375113dfec0c741e0c",
+    }
+
+    def _write_golden(self) -> IndexedTarStore:
+        store = IndexedTarStore(self.itar_path, mode="w")
+        root = create_root_group(store, {"some": "thing", "nested": {"a": [1, 2.5, None]}})
+        root.create_array(
+            "foo",
+            np.arange(27, dtype=np.float64).reshape(3, 3, 3),
+            chunks=(2, 3, 3),
+            compression=nodes.BloscCompression(shuffle="shuffle"),
+        )
+        sub = root.create_group("subgroup", attributes={"other": [1, 2, 3]})
+        sub.create_array("bar", np.arange(10, dtype=np.uint16), compression=None)
+        sub.create_bytes_array("blob", b"\x89PNG\x00data")
+        consolidate_store(store)
+        close_store(store)
+        return IndexedTarStore(self.itar_path)
+
+    @unittest.skipIf(ZARR_PYTHON_3, "zarr-python 2 metadata serialization")
+    def test_archives_of_ncore_19_8(self):
+        """Records are byte-identical to records written by ncore 19.8, archives are deterministic (node metadata
+        records are appended when closing archives)"""
+        store = self._write_golden()
+        self.assertEqual({key: hashlib.sha256(store[key]).hexdigest() for key in store}, self.GOLDEN_RECORDS)
+        # data records in creation order, followed by node metadata records (parents first)
+        self.assertEqual(
+            store.keys(),
+            [
+                "foo/0.0.0",
+                "foo/1.0.0",
+                "subgroup/bar/0",
+                "subgroup/blob/0",
+                ".zmetadata.cbor.xz",
+                ".zattrs",
+                ".zgroup",
+                "foo/.zarray",
+                "subgroup/.zattrs",
+                "subgroup/.zgroup",
+                "subgroup/bar/.zarray",
+                "subgroup/blob/.zarray",
+            ],
+        )
+        digest = hashlib.sha256(Path(self.itar_path).read_bytes()).hexdigest()
+        self.assertIn(
+            digest,
+            {
+                # (python's tarfile module encodes the device numbers of tar headers differently before python 3.9)
+                "f2f98df32a018a2f178f3561e21bb14eb3bba9da51e00956bb09ed53177000ea",  # python 3.8
+                "10c02f6fdbaad3c048af23c47eb50fcb539d660d1487df6c42c6e14c14b98a13",  # python >= 3.9
+            },
+        )
+
+    def test_archives_equivalent_to_ncore_19_8(self):
+        """Chunk records are byte-identical to chunk records written by ncore 19.8, node metadata documents are
+        equivalent (zarr-python 3 serializes them differently, and stores empty attributes of arrays)"""
+        store = self._write_golden()
+        # (zarr-python 3 writes the chunks of an array concurrently, in any order)
+        chunks = sorted(key for key in store if not is_node_metadata_key(key) and key != ".zmetadata.cbor.xz")
+        self.assertEqual(chunks, ["foo/0.0.0", "foo/1.0.0", "subgroup/bar/0", "subgroup/blob/0"])
+        self.assertEqual(
+            {key: hashlib.sha256(store[key]).hexdigest() for key in chunks},
+            {key: self.GOLDEN_RECORDS[key] for key in chunks},
+        )
+
+        metadata = decode_cbor_xz(store[".zmetadata.cbor.xz"])["metadata"]
+        self.assertEqual(metadata, {key: json.loads(store[key]) for key in store if is_node_metadata_key(key)})
+        self.assertEqual(metadata[".zattrs"], {"some": "thing", "nested": {"a": [1, 2.5, None]}})
+        self.assertEqual(metadata["subgroup/.zattrs"], {"other": [1, 2, 3]})
+        self.assertEqual(metadata.get("foo/.zattrs", {}), {})
+        self.assertEqual(
+            {key: value for key, value in metadata["foo/.zarray"].items() if key != "dimension_separator"},
+            {
+                "shape": [3, 3, 3],
+                "chunks": [2, 3, 3],
+                "dtype": "<f8",
+                "fill_value": 0.0,
+                "order": "C",
+                "filters": None,
+                "compressor": {"id": "blosc", "cname": "lz4", "clevel": 5, "shuffle": 1, "blocksize": 0},
+                "zarr_format": 2,
+            },
+        )
+        self.assertEqual(metadata["subgroup/blob/.zarray"]["dtype"], "|S9")
+        self.assertEqual(metadata["subgroup/blob/.zarray"]["shape"], [])
+
+    @parameterized.parameterized.expand([("itar",), ("directory",)])
+    def test_chunks_of_fill_values_are_stored(self, store_type: str):
+        """Chunks equal to the fill value are stored (zarr-python 3 omits them by default, zarr-python 2 doesn't)"""
+        root = create_root_group(store := self._store(store_type, "w"), {})
+        root.create_array("zeros", np.zeros((4, 3), dtype=np.float32), chunks=(2, 3))
+        root.create_bytes_array("blob", b"\x00\x00")
+        consolidate_store(store)
+        close_store(store)
+
+        if store_type == "itar":
+            keys = set(IndexedTarStore(self.itar_path).keys())
+        else:
+            keys = {
+                str(p.relative_to(self.directory_path)) for p in Path(self.directory_path).rglob("*") if p.is_file()
+            }
+        self.assertLessEqual({"zeros/0.0", "zeros/1.0", "blob/0"}, keys)
+
+    @parameterized.parameterized.expand([("itar",), ("directory",)])
+    def test_zarr_format_v3_layout(self, store_type: str):
+        """The root zarr.json of consolidated zarr format v3 hierarchies is stored compressed only (CBOR + zlib, including
+        the consolidated metadata), all other node metadata records as written by zarr-python"""
+        _require(self, 3)
+        _write(self._store(store_type, "w"), zarr_format=3)
+        if store_type == "itar":
+            records = {key: IndexedTarStore(self.itar_path)[key] for key in IndexedTarStore(self.itar_path)}
+        else:
+            root_path = Path(self.directory_path)
+            records = {str(p.relative_to(root_path)): p.read_bytes() for p in root_path.rglob("*") if p.is_file()}
+
+        self.assertIn(V3_CONSOLIDATED_METADATA_KEY, records)
+        self.assertNotIn(V2_CONSOLIDATED_METADATA_KEY, records)
+        self.assertFalse(any(key.endswith((".zarray", ".zgroup", ".zattrs")) for key in records))
+        node_documents = {
+            key[: -len("/zarr.json")]: json.loads(value) for key, value in records.items() if key.endswith("/zarr.json")
+        }
+        self.assertEqual(sorted(node_documents), ["foo", "subgroup", "subgroup/blob", "subgroup/empty", "subgroup/foo"])
+
+        root = _decode_v3_root(records[V3_CONSOLIDATED_METADATA_KEY])
+        self.assertEqual((root["zarr_format"], root["node_type"], root["attributes"]), (3, "group", _ROOT_ATTRIBUTES))
+        consolidated = root["consolidated_metadata"]
+        self.assertEqual((consolidated["kind"], consolidated["must_understand"]), ("inline", False))
+        self.assertEqual(consolidated["metadata"], node_documents)
+
+        # binary blobs are uint8 arrays, compression is the blosc codec of zarr format v3
+        self.assertEqual(
+            (node_documents["subgroup/blob"]["data_type"], node_documents["subgroup/blob"]["shape"]), ("uint8", [21])
+        )
+        self.assertEqual([codec["name"] for codec in node_documents["subgroup/foo"]["codecs"]], ["bytes", "blosc"])
+        self.assertEqual(
+            {k: v for k, v in node_documents["subgroup/foo"]["codecs"][1]["configuration"].items() if k != "typesize"},
+            {"cname": "lz4", "clevel": 5, "shuffle": "bitshuffle", "blocksize": 0},
+        )
+
+    def test_zarr_format_v3_requires_zarr_python_3(self):
+        if ZARR_PYTHON_3:
+            self.skipTest("zarr-python 2 only")
+        with self.assertRaisesRegex(ValueError, "zarr format v3 is not supported"):
+            create_root_group(self._store("itar", "w"), {}, zarr_format=3)
+
+    def test_unsupported_zarr_formats(self):
+        with self.assertRaisesRegex(ValueError, "zarr format v4 is not supported"):
+            create_root_group(self._store("directory", "w"), {}, zarr_format=4)
+
+    def test_empty(self):
+        """Stores without a root group can be consolidated, but not opened"""
+        store = IndexedTarStore(self.itar_path, mode="w")
+        consolidate_store(store)
+        close_store(store)
+        with self.assertRaises(Exception):  # (zarr-python's error of missing groups)
+            open_store(IndexedTarStore(self.itar_path), consolidated=True)
+
+    @parameterized.parameterized.expand([(zarr_format,) for zarr_format in ZARR_FORMATS])
+    def test_read_only(self, zarr_format: int):
+        _require(self, zarr_format)
+        _write(self._store("itar", "w"), zarr_format=zarr_format)
+        root = open_store(IndexedTarStore(self.itar_path))
+        with self.assertRaises(Exception):
+            root.create_group("new")
+
+
+class TestHierarchies(unittest.TestCase):
+    """Hierarchies created in arbitrary order (e.g., by multiple component writers)"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        self.path = os.path.join(self._tmp.name, "store.itar")
+
+    def tearDown(self):
+        self._tmp.cleanup()
 
     @parameterized.parameterized.expand(
-        itertools.product(
-            COMPRESSED_CONSOLIDATED_VALUES,
-            INDEX_TAIL_READ_SIZES,
+        [(consolidated, zarr_format) for consolidated in (True, False) for zarr_format in ZARR_FORMATS]
+    )
+    def test_interleaved_hierarchy(self, consolidated: bool, zarr_format: int):
+        _require(self, zarr_format)
+        store = IndexedTarStore(self.path, mode="w")
+        root = create_root_group(store, {}, zarr_format=zarr_format)
+        a, b = root.create_group("a"), root.create_group("b")
+        for i in range(3):  # members of a and b created alternately
+            a.create_array(f"x{i}", np.full(2, i))
+            b.create_group(f"g{i}").create_array("y", np.full(2, 10 + i))
+        root.create_array("z", np.zeros(1))
+        consolidate_store(store)
+        close_store(store)
+
+        root = open_store(IndexedTarStore(self.path), consolidated=consolidated)
+        self.assertEqual(sorted(root.members()), ["a", "b", "z"])
+        self.assertEqual(sorted(root.group("a").members()), ["x0", "x1", "x2"])
+        self.assertEqual(sorted(root.group("b").members()), ["g0", "g1", "g2"])
+        for i in range(3):
+            np.testing.assert_array_equal(root.array(f"a/x{i}").read(), [i, i])
+            np.testing.assert_array_equal(root.group("b").group(f"g{i}").array("y").read(), [10 + i] * 2)
+
+
+@unittest.skipUnless(ZARR_PYTHON_3, "zarr-python 3 backend")
+class TestZarr3Backend(unittest.TestCase):
+    """Execution of zarr-python 3 coroutines, and the zarr-python 3 store of indexed tar archives"""
+
+    def test_coroutines_run_on_the_calling_thread(self):
+        from ncore.impl.data import _zarr3
+
+        async def thread() -> int:
+            return threading.get_ident()
+
+        def run_in_thread() -> int:
+            return _zarr3.run(thread())
+
+        self.assertEqual(_zarr3.run(thread()), threading.get_ident())
+        with concurrent.futures.ThreadPoolExecutor(2) as pool:
+            for future in [pool.submit(run_in_thread) for _ in range(4)]:
+                self.assertNotEqual(future.result(), threading.get_ident())
+
+        # offloaded work (e.g., of codecs) runs inline
+        async def offloaded() -> int:
+            return await asyncio.get_running_loop().run_in_executor(None, threading.get_ident)
+
+        self.assertEqual(_zarr3.run(offloaded()), threading.get_ident())
+
+    def test_coroutines_within_event_loops(self):
+        from ncore.impl.data import _zarr3
+
+        async def value() -> int:
+            return 42
+
+        async def main() -> int:
+            return _zarr3.run(value())  # (runs on zarr-python's I/O loop)
+
+        self.assertEqual(asyncio.run(main()), 42)
+
+    @unittest.skipUnless(hasattr(os, "fork"), "requires fork")
+    def test_forked_processes(self):
+        """Forked processes don't reuse the event loop of the forking thread (its selector is shared with the parent)"""
+        from ncore.impl.data import _zarr3
+
+        async def loop_and_pid() -> Tuple[int, int]:
+            return id(asyncio.get_running_loop()), os.getpid()
+
+        parent_loop, _ = _zarr3.run(loop_and_pid())  # (creates the loop of this thread)
+        read_fd, write_fd = os.pipe()
+        if (pid := os.fork()) == 0:  # child (continues on this thread)
+            try:
+                child_loop, child_pid = _zarr3.run(loop_and_pid())
+                os.write(write_fd, b"ok" if child_loop != parent_loop and child_pid == os.getpid() else b"reused")
+            finally:
+                os._exit(0)
+        os.close(write_fd)
+        result = os.read(read_fd, 16)
+        os.close(read_fd)
+        os.waitpid(pid, 0)
+        self.assertEqual(result, b"ok")
+        self.assertEqual(_zarr3.run(loop_and_pid())[0], parent_loop)  # (the parent keeps its loop)
+
+        # the parent's loop is still woken by other threads (children don't close the selector shared with the parent)
+        async def woken_by_thread() -> int:
+            loop, future = asyncio.get_running_loop(), asyncio.get_running_loop().create_future()
+            timer = threading.Timer(0.05, lambda: loop.call_soon_threadsafe(future.set_result, 42))
+            timer.start()
+            return await asyncio.wait_for(future, timeout=5)
+
+        self.assertEqual(_zarr3.run(woken_by_thread()), 42)
+
+    def test_byte_ranges(self):
+        from ncore.impl.data import _zarr3
+        from ncore.impl.data._zarr3 import (
+            OffsetByteRequest,
+            RangeByteRequest,
+            SuffixByteRequest,
+            default_buffer_prototype,
+        )
+
+        with tempfile.TemporaryDirectory() as tmp:
+            path = os.path.join(tmp, "store.itar")
+            with IndexedTarStore(path, mode="w") as store:
+                store["r"] = b"0123456789"
+            tar_store = _zarr3._TarStore(IndexedTarStore(path))
+
+            def get(byte_range: Any) -> Optional[bytes]:
+                value = _zarr3.run(tar_store.get("r", default_buffer_prototype(), byte_range))
+                return None if value is None else value.to_bytes()
+
+            self.assertEqual(get(None), b"0123456789")
+            self.assertEqual(get(RangeByteRequest(2, 5)), b"234")
+            self.assertEqual(get(RangeByteRequest(8, 20)), b"89")
+            self.assertEqual(get(OffsetByteRequest(7)), b"789")
+            self.assertEqual(get(SuffixByteRequest(3)), b"789")
+            self.assertEqual(get(SuffixByteRequest(30)), b"0123456789")
+            self.assertIsNone(_zarr3.run(tar_store.get("missing", default_buffer_prototype())))
+
+
+@unittest.skipUnless(ZARR_PYTHON_3, "zarr-python 3 backend")
+class TestDirectReads(unittest.TestCase):
+    """Direct chunk reads of arrays of indexed tar archives (zarr-python 3) are identical to zarr-python's reads"""
+
+    #: Arrays (dtype, shape, chunks, compression, missing chunks) of all supported kinds
+    ARRAYS = {
+        "f4_multi": (np.float32, (7, 5, 3), (3, 2, 3), nodes.BloscCompression(), ()),
+        "u2_single_uncompressed": (np.uint16, (11,), (11,), None, ()),
+        "i8_multi_uncompressed": (np.int64, (9, 4), (4, 3), None, ()),
+        "bool": (np.bool_, (6, 6), (4, 4), nodes.BloscCompression(cname="zstd", shuffle="noshuffle"), ()),
+        "u1_partial_last": (np.uint8, (13,), (5,), nodes.BloscCompression(shuffle="shuffle"), ()),
+        "f8_missing": (np.float64, (6, 2), (2, 2), nodes.BloscCompression(), ("1.0", "c/1/0")),
+        "f4_empty": (np.float32, (0, 3), (1, 3), nodes.BloscCompression(), ()),
+    }
+
+    SELECTIONS = [
+        Ellipsis,
+        (),
+        0,
+        -1,
+        (slice(1, 4),),
+        (slice(None), 1),
+        (2, slice(0, 2)),
+        (slice(2, 2),),
+        (Ellipsis, 0),
+        (slice(-3, None),),
+        (0, slice(None), slice(1, 2)),  # (dropped axes next to selected single elements)
+        (slice(0, 3), slice(0, 2), 1),  # (complete chunks, not complete arrays)
+    ]
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def _write(self, zarr_format: int, skip_records: Tuple[str, ...] = ()) -> str:
+        """Writes all arrays (chunks of `missing chunks` dropped from the archive) and returns the archive path"""
+        rng = np.random.default_rng(1)
+        path = os.path.join(self._tmp.name, f"v{zarr_format}.itar")
+        full = os.path.join(self._tmp.name, f"v{zarr_format}.full.itar")
+        root = create_root_group(store := IndexedTarStore(full, mode="w"), {}, zarr_format=zarr_format)
+        for name, (dtype, shape, chunks, compression, _) in self.ARRAYS.items():
+            data = rng.random(shape) > 0.5 if dtype == np.bool_ else (rng.random(shape) * 1000).astype(dtype)
+            root.create_array(name, data, chunks=chunks, compression=compression)
+        root.create_bytes_array("blob", b"\x89PNG\x00binary")
+        root.create_bytes_array("blob_missing", b"some bytes")
+        consolidate_store(store)
+        close_store(store)
+
+        # copy without the missing chunks (stores of hierarchies written partially, e.g. by other writers)
+        source = IndexedTarStore(full)
+        missing = {f"{name}/{key}" for name, spec in self.ARRAYS.items() for key in spec[4]}
+        missing |= {"blob_missing/0", "blob_missing/c/0"}
+        with IndexedTarStore(path, mode="w") as copy:
+            for key in source.keys():
+                if key not in missing:
+                    copy[key] = source[key]
+        return path
+
+    def _reads(self, path: str) -> Dict[str, Any]:
+        root = open_store(IndexedTarStore(path))
+        out: Dict[str, Any] = {}
+        for name, (_, shape, *_rest) in self.ARRAYS.items():
+            array = root.array(name)
+            for selection in self.SELECTIONS:
+                try:
+                    out[f"{name}[{selection!r}]"] = array.read(selection)
+                except IndexError as e:  # (e.g., integer selections of empty arrays)
+                    out[f"{name}[{selection!r}]"] = type(e)
+        out["blob"] = root.array("blob").read_bytes()
+        # (fill value of missing chunks; zarr-python returns empty fixed-length bytes of dtype |S0)
+        out["blob_missing"] = root.array("blob_missing").read_bytes()
+        return out
+
+    @parameterized.parameterized.expand([(zarr_format,) for zarr_format in ZARR_FORMATS])
+    def test_direct_reads_match_zarr(self, zarr_format: int):
+        from ncore.impl.data import _zarr3
+
+        path = self._write(zarr_format)
+        with mock.patch.object(_zarr3, "_read_direct", wraps=_zarr3._read_direct) as direct:
+            actual = self._reads(path)
+        self.assertGreater(direct.call_count, len(actual) - 10)  # (direct reads were used)
+        with mock.patch.dict(os.environ, {_zarr3.DIRECT_READS_ENV: "0"}):
+            with mock.patch.object(_zarr3, "_read_direct", side_effect=AssertionError("direct read")):
+                expected = self._reads(path)
+
+        self.assertEqual(sorted(actual), sorted(expected))
+        for key, value in expected.items():
+            if isinstance(value, type):
+                self.assertIs(actual[key], value, key)
+                continue
+            actual_value = actual[key]
+            if isinstance(value, bytes):
+                self.assertEqual(actual_value, value, key)
+                continue
+            value = np.asarray(value)
+            self.assertEqual((actual_value.dtype, actual_value.shape), (value.dtype, value.shape), key)
+            np.testing.assert_array_equal(actual_value, value, key)
+            self.assertTrue(actual_value.flags.writeable, key)
+
+    def test_missing_chunks_of_fixed_length_bytes(self):
+        """Missing chunks of fixed-length bytes read as the fill value (empty bytes), as with zarr-python"""
+        root = open_store(IndexedTarStore(self._write(2)))
+        self.assertEqual(root.array("blob_missing").read_bytes(), b"")
+
+    def test_unsupported_selections_use_zarr(self):
+        from ncore.impl.data import _zarr3
+
+        root = open_store(IndexedTarStore(self._write(3)))
+        array = root.array("f4_multi")
+        with mock.patch.object(_zarr3, "_read_direct", side_effect=AssertionError("direct read")):
+            expected = array.read((slice(0, 6, 2),))  # (non-unit step)
+        np.testing.assert_array_equal(expected, array.read()[0:6:2])
+
+    def test_partial_chunk_selections(self):
+        """Selections of parts of compressed chunks of the output array (no in-place decompression)"""
+        root = open_store(IndexedTarStore(self._write(2)))
+        data = root.array("f4_multi").read()
+        for selection in [(slice(1, 3), slice(0, 2)), (slice(0, 3), slice(0, 2), slice(0, 2)), (slice(3, 7),)]:
+            np.testing.assert_array_equal(root.array("f4_multi").read(selection), data[selection], str(selection))
+
+    def test_stores_opened_for_writing_use_zarr(self):
+        """(node metadata of archives opened for writing is not final)"""
+        from ncore.impl.data import _zarr3
+
+        root = create_root_group(IndexedTarStore(os.path.join(self._tmp.name, "w.itar"), mode="w"), {})
+        self.assertIsNone(_zarr3._direct_read(root.create_array("a", np.arange(4)).zarr))
+
+    def test_non_native_byte_order_uses_zarr(self):
+        from ncore.impl.data import _zarr3
+
+        path = os.path.join(self._tmp.name, "be.itar")
+        data = np.arange(6, dtype=">u4")
+        root = create_root_group(store := IndexedTarStore(path, mode="w"), {})
+        root.create_array("a", data)
+        consolidate_store(store)
+        close_store(store)
+        array = open_store(IndexedTarStore(path)).array("a")
+        self.assertIsNone(_zarr3._direct_read(array.zarr))
+        np.testing.assert_array_equal(array.read(), data)
+
+    def test_directory_stores_use_zarr(self):
+        from ncore.impl.data import _zarr3
+
+        root = create_root_group(store := open_directory_store(os.path.join(self._tmp.name, "dir")), {})
+        root.create_array("a", np.arange(4))
+        consolidate_store(store)
+        self.assertIsNone(
+            _zarr3._direct_read(open_store(open_directory_store(os.path.join(self._tmp.name, "dir"))).array("a").zarr)
+        )
+
+    def test_large_chunks_are_decoded_concurrently(self):
+        from ncore.impl.data import _zarr3
+
+        path = os.path.join(self._tmp.name, "large.itar")
+        data = np.arange(4 * (1 << 20), dtype=np.float32).reshape(4, -1)  # (4 chunks of 4 MiB)
+        root = create_root_group(store := IndexedTarStore(path, mode="w"), {})
+        root.create_array("a", data, chunks=(1, data.shape[1]))
+        consolidate_store(store)
+        close_store(store)
+
+        array = open_store(IndexedTarStore(path)).array("a")
+        threads: List[str] = []
+        decode = np.frombuffer
+
+        def record(*args: Any, **kwargs: Any) -> Any:
+            threads.append(threading.current_thread().name)
+            return decode(*args, **kwargs)
+
+        with mock.patch.object(_zarr3.np, "frombuffer", side_effect=record):
+            np.testing.assert_array_equal(array.read((slice(None), slice(0, 10))), data[:, :10])
+        self.assertEqual(len(threads), 4)
+        self.assertTrue(all(name.startswith("ncore_zarr3_chunks") for name in threads), threads)
+
+        # reads beyond the bound decode their chunks on the calling thread
+        threads.clear()
+        parallel = _zarr3._parallel()
+        for _ in range(_zarr3._PARALLEL_READS):
+            parallel.semaphore.acquire()
+        try:
+            with mock.patch.object(_zarr3.np, "frombuffer", side_effect=record):
+                array.read((slice(None), slice(0, 10)))
+        finally:
+            for _ in range(_zarr3._PARALLEL_READS):
+                parallel.semaphore.release()
+        self.assertEqual(threads, [threading.current_thread().name] * 4)
+
+    @unittest.skipUnless(hasattr(os, "fork"), "requires fork")
+    def test_forked_processes(self):
+        """Forked processes don't inherit the parallel reads state (its locks / threads are unusable in children)"""
+        from ncore.impl.data import _zarr3
+
+        parent = _zarr3._parallel()
+        read_fd, write_fd = os.pipe()
+        if (pid := os.fork()) == 0:
+            try:
+                os.write(write_fd, b"ok" if _zarr3._parallel() is not parent else b"inherited")
+            finally:
+                os._exit(0)
+        os.close(write_fd)
+        result = os.read(read_fd, 16)
+        os.close(read_fd)
+        os.waitpid(pid, 0)
+        self.assertEqual(result, b"ok")
+        self.assertIs(_zarr3._parallel(), parent)
+
+
+class TestAsyncReads(unittest.TestCase):
+    """Asynchronous reads return the same data as synchronous reads, without blocking the running event loop"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        self.zarr_format = max(nodes.SUPPORTED_ZARR_FORMATS)
+        self.path = os.path.join(self._tmp.name, "store.itar")
+        _write(IndexedTarStore(self.path, mode="w"), zarr_format=self.zarr_format)
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def _reads(self, root: nodes.Group) -> List[Any]:
+        return [
+            root.array("foo").read(),
+            root.array("subgroup/foo").read((slice(1, 3),)),
+            root.array("subgroup/foo").read((2, slice(None), 0)),
+            root.array("subgroup/empty").read(),
+            root.array("subgroup/blob").read_bytes(),
+        ]
+
+    async def _reads_async(self, root: nodes.Group) -> List[Any]:
+        return list(
+            await asyncio.gather(
+                root.array("foo").read_async(),
+                root.array("subgroup/foo").read_async((slice(1, 3),)),
+                root.array("subgroup/foo").read_async((2, slice(None), 0)),
+                root.array("subgroup/empty").read_async(),
+                root.array("subgroup/blob").read_bytes_async(),
+            )
+        )
+
+    def _assert_equal(self, actual: List[Any], expected: List[Any]) -> None:
+        self.assertEqual(len(actual), len(expected))
+        for a, e in zip(actual, expected):
+            if isinstance(e, bytes):
+                self.assertEqual(a, e)
+            else:
+                self.assertEqual((a.dtype, a.shape), (e.dtype, e.shape))
+                np.testing.assert_array_equal(a, e)
+
+    @parameterized.parameterized.expand([("direct",), ("zarr",)])
+    def test_async_reads(self, mode: str):
+        root = open_store(IndexedTarStore(self.path))
+        expected = self._reads(root)
+        with mock.patch.dict(os.environ, {"NCORE_ZARR3_DIRECT_READS": "1" if mode == "direct" else "0"}):
+            self._assert_equal(asyncio.run(self._reads_async(root)), expected)  # (an application's event loop)
+            if ZARR_PYTHON_3:
+                from ncore.impl.data import _zarr3
+
+                self._assert_equal(_zarr3.run(self._reads_async(root)), expected)  # (the loop of the calling thread)
+
+    @parameterized.parameterized.expand([("direct",), ("zarr",)])
+    def test_async_reads_dont_block_event_loops(self, mode: str):
+        """Blocking parts of reads awaited on event loops of applications (record reads, decompression) run on
+        worker threads"""
+        root = open_store(IndexedTarStore(self.path))
+        expected = self._reads(root)
+        threads = set()
+        getitem = IndexedTarStore.__getitem__
+
+        def recording_getitem(store: IndexedTarStore, key: str) -> bytes:
+            threads.add(threading.current_thread().name)
+            return getitem(store, key)
+
+        with mock.patch.dict(os.environ, {"NCORE_ZARR3_DIRECT_READS": "1" if mode == "direct" else "0"}):
+            with mock.patch.object(IndexedTarStore, "__getitem__", recording_getitem):
+                self._assert_equal(asyncio.run(self._reads_async(root)), expected)
+        self.assertTrue(threads, "no record reads")
+        self.assertNotIn(threading.current_thread().name, threads)
+
+    @parameterized.parameterized.expand([("direct",), ("zarr",)])
+    def test_async_reads_on_the_loop_of_the_calling_thread_run_inline(self, mode: str):
+        """Reads awaited on the event loop of the calling thread (see _zarr3.run) don't hand off to threads"""
+        if not ZARR_PYTHON_3:
+            self.skipTest("event loops of the calling threads (zarr-python 3)")
+        from ncore.impl.data import _zarr3
+
+        root = open_store(IndexedTarStore(self.path))
+        expected = self._reads(root)
+        threads = set()
+        getitem = IndexedTarStore.__getitem__
+
+        def recording_getitem(store: IndexedTarStore, key: str) -> bytes:
+            threads.add(threading.current_thread().name)
+            return getitem(store, key)
+
+        with mock.patch.dict(os.environ, {"NCORE_ZARR3_DIRECT_READS": "1" if mode == "direct" else "0"}):
+            with mock.patch.object(IndexedTarStore, "__getitem__", recording_getitem):
+                self._assert_equal(_zarr3.run(self._reads_async(root)), expected)
+        self.assertEqual(threads, {threading.current_thread().name})
+
+    def test_sync_reads_within_event_loops(self):
+        """Synchronous reads also work within running event loops"""
+        root = open_store(IndexedTarStore(self.path))
+        expected = self._reads(root)
+
+        async def main() -> List[Any]:
+            return self._reads(open_store(IndexedTarStore(self.path)))
+
+        self._assert_equal(asyncio.run(main()), expected)
+
+    @unittest.skipUnless(ZARR_PYTHON_3, "range requests of remote archives (zarr-python 3)")
+    def test_remote_range_requests(self):
+        """Chunks of remote archives are read via asynchronous range requests on the event loop of their filesystem,
+        also when awaited on other event loops (e.g., fsspec's http filesystem binds its sessions to fsspec's I/O
+        loop)"""
+        import fsspec.asyn
+
+        class LoopBoundFileSystem(fsspec.asyn.AsyncFileSystem):
+            """A filesystem whose coroutines only work on its own event loop (as fsspec's http filesystem)"""
+
+            async_impl = True
+
+            async def _cat_file(self, path: str, start: Optional[int] = None, end: Optional[int] = None, **_: Any):
+                if asyncio.get_running_loop() is not self.loop:
+                    raise RuntimeError("Timeout context manager should be used inside a task")
+                requests.append((start, end))
+                with open(path, "rb") as f:
+                    f.seek(start or 0)
+                    return f.read(None if end is None else end - (start or 0))
+
+        requests: List[Tuple[Optional[int], Optional[int]]] = []
+        fs = LoopBoundFileSystem(loop=fsspec.asyn.get_loop())
+        store = IndexedTarStore(self.path, index_tail_read_size=None)  # (chunks aren't read with the index)
+        root = open_store(store)
+        expected = self._reads(root)
+        with mock.patch.object(IndexedTarStore, "_async_fs", lambda self: fs):
+            self.assertTrue(store.is_remote)
+            self._assert_equal(asyncio.run(self._reads_async(root)), expected)
+        self.assertGreaterEqual(len(requests), 4)  # (range requests of the chunks)
+        self.assertFalse(IndexedTarStore(self.path).is_remote)  # (local archives use positional reads)
+
+
+@parameterized.parameterized_class(("zarr_format",), [(zarr_format,) for zarr_format in ZARR_FORMATS])
+class TestNodes(unittest.TestCase):
+    """Typed node API"""
+
+    zarr_format: int
+
+    def setUp(self):
+        _require(self, self.zarr_format)
+        self._tmp = tempfile.TemporaryDirectory()
+        self.root = create_root_group(
+            open_directory_store(self._tmp.name),
+            {"s": "x", "i": 3, "b": True, "l": [1], "d": {"k": 1}},
+            zarr_format=self.zarr_format,
         )
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def test_typed_attributes(self):
+        self.assertEqual(self.root.attr_str("s"), "x")
+        self.assertEqual(self.root.attr_int("i"), 3)
+        self.assertEqual(self.root.attr_list("l"), [1])
+        self.assertEqual(self.root.attr_dict("d"), {"k": 1})
+        with self.assertRaises(TypeError):
+            self.root.attr_int("b")  # (booleans aren't integers)
+        with self.assertRaises(TypeError):
+            self.root.attr_str("i")
+        with self.assertRaises(KeyError):
+            self.root.attr_str("missing")
+
+        self.root.update_attrs({"i": 4, "new": None})
+        self.assertEqual(self.root.attrs, {"s": "x", "i": 4, "b": True, "l": [1], "d": {"k": 1}, "new": None})
+
+    def test_arrays(self):
+        array = self.root.create_array("a", [1.5, 2.5], dtype=np.float32, attributes={"unit": "m"})
+        self.assertEqual((array.shape, array.dtype, array.attrs), ((2,), np.dtype(np.float32), {"unit": "m"}))
+        self.assertEqual(array.compression, nodes.BloscCompression())
+        np.testing.assert_array_equal(array.read((slice(1, 2),)), [2.5])
+        self.assertEqual(array.read((0,)).shape, ())  # 0-d arrays for scalar selections
+
+        self.assertIsNone(self.root.create_array("u", np.arange(3), compression=None).compression)
+        self.assertEqual(self.root.create_array("c", np.zeros((10, 0)), chunks=(4, 0)).zarr.chunks, (4, 1))
+        with self.assertRaises(TypeError):
+            array.read_bytes()
+
+    def test_bytes_arrays(self):
+        for data in (b"\x89PNG\x00binary", b"", b"\x00"):
+            array = self.root.create_bytes_array(f"blob{len(data)}", data, attributes={"format": "png"})
+            self.assertEqual(array.zarr_format, self.zarr_format)
+            self.assertEqual(array.read_bytes(), data.rstrip(b"\x00") if self.zarr_format == 2 else data)
+            self.assertEqual(array.attrs, {"format": "png"})
+            self.assertIsNone(array.compression)
+        # (zarr format v2: fixed-length bytes, as ncore <= 19.8, which strip trailing NUL bytes)
+        self.assertEqual(self.root.array("blob11").dtype, np.dtype("S11") if self.zarr_format == 2 else np.uint8)
+
+    @parameterized.parameterized.expand(
+        [(None,), (nodes.BloscCompression(),), (nodes.BloscCompression("zstd", 3, "shuffle", 256),)]
+    )
+    def test_compression(self, compression: Optional[nodes.BloscCompression]):
+        array = self.root.create_array("a", np.arange(1000, dtype=np.int32), compression=compression)
+        self.assertEqual(array.compression, compression)
+        np.testing.assert_array_equal(array.read(), np.arange(1000))
+
+    def test_groups(self):
+        self.root.create_group("b")
+        self.root.create_group("a").create_group("nested")
+        self.assertEqual([name for name, _ in self.root.groups()], ["a", "b"])
+        self.assertEqual(self.root.require_group("a").group("nested").path, "a/nested")
+        self.assertEqual(self.root.group("a/nested").path, "a/nested")
+        with self.assertRaises(Exception):
+            self.root.create_group("a")
+
+    def test_nodes(self):
+        """Members of unknown kind (groups or arrays)"""
+        self.root.create_group("g").create_array("a", np.arange(3))
+        assert isinstance(group := self.root.node("g"), nodes.Group)
+        self.assertEqual(group.path, "g")
+        assert isinstance(array := self.root.node("g/a"), nodes.Array)
+        np.testing.assert_array_equal(array.read(), np.arange(3))
+        with self.assertRaises(KeyError):
+            self.root.node("g/missing")
+
+
+class TestNodeCache(unittest.TestCase):
+    """Node caches of groups (disabled / unbounded / least recently used)"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        path = os.path.join(self._tmp.name, "store.itar")
+        store = IndexedTarStore(path, mode="w")
+        sub = create_root_group(store, {}).create_group("g")
+        for i in range(8):
+            sub.create_array(f"a{i}", np.full(3, i))
+        consolidate_store(store)
+        close_store(store)
+        self.root = open_store(IndexedTarStore(path))
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def test_uncached(self):
+        self.assertIsNone(self.root.cache)
+        self.assertIsNot(self.root.array("g/a1").zarr, self.root.array("g/a1").zarr)
+
+    def test_unbounded(self):
+        root = self.root.with_cache(cache := nodes.NodeCache())
+        for i in range(8):
+            np.testing.assert_array_equal(root.array(f"g/a{i}").read(), [i] * 3)
+        self.assertEqual(len(cache), 8)
+        self.assertIs(root.array("g/a3").zarr, root.array("g/a3").zarr)
+
+    def test_member_groups_share_the_cache(self):
+        root = self.root.with_cache(cache := nodes.NodeCache())
+        group = root.group("g")
+        self.assertIs(group.cache, cache)
+        node = root.node("g")
+        assert isinstance(node, nodes.Group)
+        self.assertIs(node.cache, cache)
+        self.assertIs(cast(nodes.Array, root.node("g/a2")).zarr, root.array("g/a2").zarr)
+        self.assertIs(group.array("a3").zarr, root.array("g/a3").zarr)
+        self.assertIs(next(root.groups())[1].cache, cache)
+        self.assertEqual(cache.keys(), ["g", "g/a2", "g/a3"])
+
+    def test_least_recently_used(self):
+        root = self.root.with_cache(cache := nodes.NodeCache(max_entries=3))
+        a0 = root.array("g/a0").zarr
+        root.array("g/a1")
+        root.array("g/a2")
+        self.assertIs(root.array("g/a0").zarr, a0)  # (a1 is the least recently used node now)
+        root.array("g/a3")
+        self.assertEqual(cache.keys(), ["g/a2", "g/a0", "g/a3"])
+        for i in range(8):  # evicted nodes are re-opened
+            np.testing.assert_array_equal(root.array(f"g/a{i}").read(), [i] * 3)
+        self.assertEqual(len(cache), 3)
+
+    def test_missing_nodes_are_not_cached(self):
+        root = self.root.with_cache(cache := nodes.NodeCache())
+        with self.assertRaises(KeyError):
+            root.array("g/missing")
+        self.assertEqual(len(cache), 0)
+
+    def test_invalid_sizes(self):
+        with self.assertRaises(ValueError):
+            nodes.NodeCache(max_entries=0)
+
+
+class TestBloscCompression(unittest.TestCase):
+    @parameterized.parameterized.expand(
+        [({"cname": "snappy"},), ({"clevel": 10},), ({"clevel": -1},), ({"shuffle": "byte"},), ({"blocksize": -1},)]
     )
-    def test_empty(self, compressed_consolidate: bool, index_tail_read_size: Optional[int]):
-        """Verify edge case of serialization of empty store is possible without errors"""
-        with tempfile.NamedTemporaryFile(suffix=".itar") as f:
-            with IndexedTarStore(f.name, mode="w") as s_itar_out:  # closes file on exit
-                # Don't write any zarr data (still serializes empty tar / seek tables)
-
-                if compressed_consolidate:
-                    consolidate_compressed_metadata(s_itar_out)
-
-            with IndexedTarStore(f.name, index_tail_read_size=index_tail_read_size) as s_itar_in:
-                # Loading store should work without errors
-
-                # But loading a non-existing group should then fail
-                with self.assertRaises(zarr.errors.PathNotFoundError):
-                    if compressed_consolidate:
-                        open_compressed_consolidated(s_itar_in, mode="r")
-                    else:
-                        zarr.open(store=s_itar_in, mode="r")
-
-    def test_reload_resources_preserves_write_mode_archive(self):
-        """Verify writer resource reload does not truncate already-written archive data."""
-        with tempfile.NamedTemporaryFile(suffix=".itar") as f:
-            with IndexedTarStore(f.name, mode="w") as store:
-                store["first"] = b"first payload"
-                store.tar_file_object.flush()
-
-                size_before_reload = Path(f.name).stat().st_size
-                self.assertGreater(size_before_reload, 0)
-
-                store.reload_resources()
-
-                self.assertEqual(Path(f.name).stat().st_size, size_before_reload)
-                self.assertEqual(store["first"], b"first payload")
-
-                store["second"] = b"second payload"
-                self.assertEqual(store["first"], b"first payload")
-                self.assertEqual(store["second"], b"second payload")
-
-            with IndexedTarStore(f.name, mode="r") as store:
-                self.assertEqual(store["first"], b"first payload")
-                self.assertEqual(store["second"], b"second payload")
-
-    def test_index_tail_cache_avoids_extra_file_reads(self):
-        """Payloads fully covered by the index tail read must not trigger further ``read()`` calls."""
-
-        class _CountingReader(IO[Any]):
-            __slots__ = ("_f", "read_calls")
-
-            def __init__(self, f: IO[Any]):
-                self._f = f
-                self.read_calls = 0
-
-            def read(self, n=-1):
-                self.read_calls += 1
-                return self._f.read(n)
-
-            def seek(self, *args, **kwargs):
-                return self._f.seek(*args, **kwargs)
-
-            def tell(self):
-                return self._f.tell()
-
-            def close(self):
-                return self._f.close()
-
-            def __getattr__(self, name):
-                return getattr(self._f, name)
-
-        with tempfile.NamedTemporaryFile(suffix=".itar") as f:
-            g_ref = zarr.open(store=zarr.MemoryStore())
-            g_ref.create_dataset("x", data=np.array([1.0, 2.0]))
-            with IndexedTarStore(f.name, mode="w") as s_out:
-                zarr.copy_store(g_ref.store, s_out)
-
-            file_size = Path(f.name).stat().st_size
-            self.assertGreater(file_size, 0)
-            store = IndexedTarStore(f.name, index_tail_read_size=file_size)
-
-            # Wrap the file object to intercept read counts
-            inner = store.tar_file_object
-            counter = _CountingReader(inner)
-            store.tar_file_object = counter
-            self.assertEqual(counter.read_calls, 0)
-
-            # Read every key via the tail buffer and verify no extra file reads occurred
-            values = {key: store[key] for key in store}
-            self.assertEqual(counter.read_calls, 0)
-
-            # Verify actual data was served from the store
-            self.assertGreater(len(values), 0)
-            g_reload = zarr.open(store=store, mode="r")
-            self.assertIsNone(np.testing.assert_array_equal(g_ref["x"][()], g_reload["x"][()]))
+    def test_invalid_parameters(self, kwargs: Any):
+        with self.assertRaises(ValueError):
+            nodes.BloscCompression(**kwargs)
+
+    @parameterized.parameterized.expand([(shuffle,) for shuffle in ("noshuffle", "shuffle", "bitshuffle")])
+    def test_numcodecs(self, shuffle: Any):
+        compression = nodes.BloscCompression(cname="zstd", clevel=3, shuffle=shuffle, blocksize=1024)
+        self.assertEqual(nodes.BloscCompression.from_numcodecs(compression.to_numcodecs()), compression)
+
+
+def _read_member(tar: tarfile.TarFile, member: tarfile.TarInfo) -> bytes:
+    file = tar.extractfile(member)
+    assert file is not None
+    return file.read()
diff --git a/ncore/impl/data/types.py b/ncore/impl/data/types.py
index e4e29cf4..3fc7a697 100644
--- a/ncore/impl/data/types.py
+++ b/ncore/impl/data/types.py
@@ -49,7 +49,7 @@
     import numpy.typing as npt  # type: ignore[import-not-found]
 
 from ncore.impl.common.transformations import PoseGraphInterpolator, transform_bbox, transform_point_cloud
-from ncore.impl.data import util
+from ncore.impl.data import _json, util
 
 
 if sys.version_info >= (3, 11):
@@ -61,15 +61,7 @@
 
 ## JSON-like structures
 
-JsonLike = Union[
-    Dict[str, "JsonLike"],
-    List["JsonLike"],
-    str,
-    int,
-    float,
-    bool,
-    None,
-]
+JsonLike = _json.JsonLike
 
 
 ## Data classes representing stored data types
diff --git a/ncore/impl/data/v4/BUILD.bazel b/ncore/impl/data/v4/BUILD.bazel
index b8a88a24..5b8e5ae1 100644
--- a/ncore/impl/data/v4/BUILD.bazel
+++ b/ncore/impl/data/v4/BUILD.bazel
@@ -99,3 +99,20 @@ pytest_test(
         "@rules_python//python/runfiles",
     ],
 )
+
+pytest_test(
+    name = "pytest_formats",
+    srcs = ["formats_test.py"],
+    data = [
+        "@test-data-v4-trimmed//:zarr_v2",
+        "@test-data-v4-trimmed//:zarr_v3",
+    ],
+    deps = [
+        ":pylib_components",
+        "//ncore/impl/data:pylib_stores",
+        requirement("numpy"),
+        requirement("parameterized"),
+        requirement("universal_pathlib"),
+        "@rules_python//python/runfiles",
+    ],
+)
diff --git a/ncore/impl/data/v4/components.py b/ncore/impl/data/v4/components.py
index bf95083d..fe160980 100644
--- a/ncore/impl/data/v4/components.py
+++ b/ncore/impl/data/v4/components.py
@@ -44,17 +44,13 @@
 import dataclasses_json
 import numpy as np
 import PIL.Image as PILImage
-import zarr
-import zarr.storage
 
-from numcodecs import Blosc
 from typing_extensions import Concatenate, ParamSpec
 from upath import UPath
-from zarr._storage.store import Store
 
 from ncore.impl.common.transformations import HalfClosedInterval
 from ncore.impl.common.util import MD5Hasher, unpack_optional
-from ncore.impl.data import stores, types, util
+from ncore.impl.data import nodes, stores, types, util
 from ncore.impl.data.types import PointCloud
 
 
@@ -72,32 +68,30 @@
 VERSION = "v4"
 
 
-def _normalize_chunks(
-    shape: "Tuple[int, ...]",
-    *,
-    require_nonzero_dims: "Optional[Tuple[int, ...]]" = None,
-) -> "Tuple[int, ...]":
-    """Normalize a chunk shape by replacing zero dimensions with 1.
+def _check_nonzero_dims(shape: Tuple[int, ...], dims: Tuple[int, ...], name: str) -> None:
+    """Asserts that the given dimensions of the shape of the named data are non-zero.
 
-    Zarr v2 rejects chunk tuples containing zeros (division by zero in internal
-    bookkeeping). This helper ensures that every dimension in the chunk tuple is
-    at least 1 while the *logical* data shape passed to ``create_dataset`` via
-    the ``data=`` argument remains unchanged.
+    Other dimensions are allowed to be zero (arrays with zero-sized dimensions are supported).
 
     Args:
-        shape: The data shape to derive a chunk shape from.
-        require_nonzero_dims: Optional tuple of dimension indices that MUST be
-            non-zero (raises AssertionError otherwise). When ``None`` (the
-            default) all dimensions are allowed to be zero and get clamped to 1
-            in the returned chunk shape.
-
-    Returns:
-        A chunk-safe tuple where any zero dimension is replaced with 1.
+        shape: The shape of the data.
+        dims: Indices of the dimensions that MUST be non-zero (raises AssertionError otherwise).
+        name: Name of the data (for error messages).
     """
-    if require_nonzero_dims is not None:
-        for d in require_nonzero_dims:
-            assert shape[d] != 0, f"Dimension {d} must be non-zero, got shape {shape}"
-    return tuple(max(1, s) for s in shape)
+    for d in dims:
+        assert shape[d] != 0, f"Dimension {d} of {name} must be non-zero, got shape {shape}"
+
+
+def _frame_id(
+    # end-of-frame timestamp, or start-of-frame / end-of-frame timestamps
+    timestamps_us: Union[int, npt.NDArray[np.uint64]],
+) -> int:
+    """Returns the frame ID (the end-of-frame timestamp) of a frame"""
+
+    if isinstance(timestamps_us, np.ndarray):
+        return cast(np.ndarray, timestamps_us)[1].item()  # end-of-frame timestamp is frame ID
+
+    return timestamps_us
 
 
 @dataclass
@@ -142,6 +136,7 @@ def from_reader(
         store_base_name: str,
         sequence_reader: SequenceComponentGroupsReader,
         store_type: Literal["itar", "directory"] = "itar",  # valid values: ['itar', 'directory']
+        zarr_format: Literal[2, 3] = 2,
     ) -> SequenceComponentGroupsWriter:
         """Creates a SequenceComponentGroupsWriter from an existing SequenceComponentGroupsReader instance to share consistent per-sequence meta-data"""
 
@@ -152,6 +147,7 @@ def from_reader(
             sequence_timestamp_interval_us=sequence_reader.sequence_timestamp_interval_us,
             generic_meta_data=sequence_reader.generic_meta_data,
             store_type=store_type,
+            zarr_format=zarr_format,
         )
 
     def __init__(
@@ -166,6 +162,9 @@ def __init__(
         generic_meta_data: Dict[str, types.JsonLike],
         # Zarr store type: either serialize as .itar archive store (default / production) or plain "directory" store (simpler for introspection / asynchronous / external setup)
         store_type: Literal["itar", "directory"] = "itar",  # valid values: ['itar', 'directory']
+        # On-disk zarr format of the component stores: format v2 (default) is readable with all zarr-python versions
+        # (and all ncore versions), format v3 requires zarr-python 3 for writing *and* reading
+        zarr_format: Literal[2, 3] = 2,
     ):
         """
         Instantiate sequence component groups writer and initialize the default data groups and file stores for a given sequence and sensor IDs
@@ -180,9 +179,10 @@ def __init__(
 
         # Individual stores for each group are initialized lazily on-demand (indexed tar file or zarr directories)
         self._stores_rootgroups: dict[
-            str, Tuple[zarr.Group, UPath]
+            str, Tuple[nodes.Group, UPath]
         ] = {}  # maps component group names to stores, store path, and base groups
         self._store_type = store_type
+        self._zarr_format = nodes.check_zarr_format(zarr_format)
 
         # registered component writers
         self._component_writers: dict[
@@ -201,7 +201,12 @@ def sequence_timestamp_interval_us(self) -> HalfClosedInterval:
     def generic_meta_data(self) -> Dict[str, types.JsonLike]:
         return self._generic_meta_data
 
-    def get_base_group(self, component_group_name: Optional[str]) -> zarr.Group:
+    @property
+    def zarr_format(self) -> int:
+        """The on-disk zarr format of the written component stores"""
+        return self._zarr_format
+
+    def get_base_group(self, component_group_name: Optional[str]) -> nodes.Group:
         """Lazily initializes ncore base-groups and underlying stores on demand"""
 
         if component_group_name is None:
@@ -209,7 +214,7 @@ def get_base_group(self, component_group_name: Optional[str]) -> zarr.Group:
             component_group_name = ""
 
         if (store_rootgroup := self._stores_rootgroups.get(component_group_name)) is not None:
-            # Store already initialized, return it's root group
+            # Store already initialized, return its root group
             return store_rootgroup[0]
 
         # Store doesn't exist yet, create it
@@ -221,23 +226,21 @@ def get_base_group(self, component_group_name: Optional[str]) -> zarr.Group:
             # append group name as suffix to store name if given
             store_name += f"-{component_group_name}"
 
-        store: Store
+        store: stores.StoreLike
         if self._store_type == "itar":
             # container-based zarr stores <base-name>.<store_name>.zarr.itar
             store_path = self._output_dir_path / f"{self._store_base_name}.{store_name}.zarr.itar"
             store = stores.IndexedTarStore(store_path, mode="w")
         elif self._store_type == "directory":
-            # directory-based zarr stores <base-name>.<store_name>.zarr.zarr
+            # directory-based zarr stores <base-name>.<store_name>.zarr
             store_path = self._output_dir_path / f"{self._store_base_name}.{store_name}.zarr"
-            store = zarr.storage.DirectoryStore(store_path)
+            store = stores.open_directory_store(store_path)
         else:
             raise ValueError(f"Unknown store type {self._store_type}")
 
-        # Create root group in store
-        root_group = zarr.group(store=store)
-
-        # Store dataset associated meta-data to root
-        root_group.attrs.put(
+        # Create root group in store, storing dataset associated meta-data to root
+        root_group = stores.create_root_group(
+            store,
             {
                 "sequence_id": self._sequence_id,
                 "sequence_timestamp_interval_us": {
@@ -247,7 +250,8 @@ def get_base_group(self, component_group_name: Optional[str]) -> zarr.Group:
                 "generic_meta_data": self._generic_meta_data,
                 "version": VERSION,
                 "component_group_name": component_group_name,
-            }
+            },
+            zarr_format=self._zarr_format,
         )
 
         # Create store / base-group mapping
@@ -267,32 +271,27 @@ def finalize(self) -> List[UPath]:
 
         # Write deferred component metadata and generic data of all writers
         for component_writer in self._component_writers.values():
-            # Merge generic_meta_data with regular component's meta data
-            (cw_group := component_writer._group).attrs.put(
+            # Merge generic_meta_data with regular component's meta data (component groups are created without
+            # attributes by register_component_writer(), so this sets the complete component attributes)
+            (cw_group := component_writer._group).update_attrs(
                 {**component_writer._component_meta_data, "generic_meta_data": component_writer._generic_meta_data}
             )
 
             # Write generic data arrays if any
             if component_writer._generic_data:
-                compressor = Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE)
                 gd_group = cw_group.require_group("generic_data")
                 for name, array in component_writer._generic_data.items():
-                    gd_group.create_dataset(
-                        name,
-                        data=array,
-                        chunks=_normalize_chunks(array.shape),
-                        compressor=compressor,
-                    )
+                    gd_group.create_array(name, array)
 
         # Make sure the stores are consolidated and closed
         ret = []
         for root_group, store_path in self._stores_rootgroups.values():
-            store = root_group.store
+            store = stores.group_store(root_group)
 
-            stores.consolidate_compressed_metadata(store)
+            stores.consolidate_store(store)
 
             # Finish writing all files
-            store.close()
+            stores.close_store(store)
 
             ret.append(store_path)
 
@@ -300,7 +299,7 @@ def finalize(self) -> List[UPath]:
 
     def register_component_writer(
         self,
-        component_writer_type: "Callable[Concatenate[zarr.Group, HalfClosedInterval, P], CW]",
+        component_writer_type: "Callable[Concatenate[nodes.Group, HalfClosedInterval, P], CW]",
         component_instance_name: str,
         group_name: Optional[str] = None,
         generic_meta_data: Optional[Dict[str, types.JsonLike]] = None,
@@ -387,6 +386,7 @@ def __init__(
         open_consolidated: bool = True,
         itar_index_tail_read_size: int = 1 << 20,  # 1 MiB default
         max_threads: int | None = None,
+        node_cache_size: int | None = None,
     ):
         """Initialize a SequenceComponentReader for a virtual sequence represented by a list of components.
 
@@ -402,28 +402,39 @@ def __init__(
                                single I/O call to initialize the tar file index to minimize reads.
                                Default 1 MiB size fits typical use cases, but may be increased if the compressed
                                index size is expected to be larger or decreased if remote access chunk sizes are smaller.
-            max_threads:       The maximum number of threads used to load the different components (if None,
-                               use interpreter-default number of threads for a ThreadPoolExecutor)
+            max_threads:       The maximum number of threads used to open the component stores concurrently (if
+                               None, use interpreter-default number of threads for a ThreadPoolExecutor)
+            node_cache_size:   Number of opened groups / arrays (with their decoded meta-data, but not their data)
+                               each component reader caches for repeated accesses (e.g., of frames). None caches
+                               all accessed nodes for the lifetime of the reader, a positive number the most recently
+                               used nodes only, and 0 disables caching.
         """
 
+        if node_cache_size is not None and node_cache_size < 0:
+            raise ValueError(f"node_cache_size needs to be None or non-negative, got {node_cache_size}")
+        self._node_cache_size = node_cache_size
+
         component_group_upaths: List[UPath] = self.expand_component_group_paths(component_group_paths)
 
         assert len(component_group_upaths), "No component inputs provided"
 
         # Load component stores concurrently (to hide latency) and check for sequence consistency
-        self._component_stores: Dict[str, Tuple[zarr.Group, UPath]] = {}  # use str as the generic path / URL type
+        self._component_stores: Dict[str, Tuple[nodes.Group, UPath]] = {}  # use str as the generic path / URL type
 
+        # Component stores are opened concurrently (hiding the latency of remote archive index reads), their
+        # hierarchies are opened from their consolidated metadata sequentially (pure-Python CPU work that threads
+        # only slow down)
         with concurrent.futures.ThreadPoolExecutor(max_workers=max_threads) as executor:
 
-            def thread_load_component_store(component_store_upath: UPath) -> Tuple[zarr.Group, UPath]:
-                """Thread-executed shard opening"""
+            def thread_open_component_store(component_store_upath: UPath) -> Tuple[stores.StoreLike, UPath]:
+                """Thread-executed component store opening"""
 
                 # Make sure paths are absolute at this point
                 component_store_upath = component_store_upath.absolute()
 
                 _logger.info(f"SequenceStoreReader: Loading component store {component_store_upath}")
 
-                component_store: Store
+                component_store: stores.StoreLike
                 if component_store_upath.is_file():
                     if not component_store_upath.name.endswith(".zarr.itar"):
                         # not a supported file-based store format
@@ -435,61 +446,50 @@ def thread_load_component_store(component_store_upath: UPath) -> Tuple[zarr.Grou
                         component_store_upath, mode="r", index_tail_read_size=itar_index_tail_read_size
                     )
                 else:
-                    component_store = zarr.storage.DirectoryStore(component_store_upath)
-
-                component_root = (
-                    stores.open_compressed_consolidated(store=component_store, mode="r")
-                    if open_consolidated
-                    else zarr.open(store=component_store, mode="r")
-                )
+                    component_store = stores.open_directory_store(component_store_upath)
 
-                return cast(zarr.Group, component_root), component_store_upath
+                return component_store, component_store_upath
 
-            for future in concurrent.futures.as_completed(
-                [
-                    executor.submit(thread_load_component_store, component_store_upath)
-                    for component_store_upath in component_group_upaths
-                ]
-            ):
-                # Note: thread completion order is not relevant here
-                component_root, component_store_path = future.result()
+            opened_component_stores = list(executor.map(thread_open_component_store, component_group_upaths))
 
-                component_root_attrs = dict(component_root.attrs.items())
+        for component_store, component_store_path in opened_component_stores:
+            component_root = stores.open_store(component_store, open_consolidated)
 
-                # Check sequence compatibility
-                component_root_version = component_root_attrs["version"]
-                if component_root_version not in [VERSION]:
-                    raise RuntimeError(
-                        f"Can't load V4 component store {component_store_path} with incompatible data version {component_root_version}"
-                    )
-
-                component_store_sequence_id = component_root_attrs["sequence_id"]
-                component_store_sequence_timestamp_interval_us = HalfClosedInterval(
-                    component_root_attrs["sequence_timestamp_interval_us"]["start"],
-                    component_root_attrs["sequence_timestamp_interval_us"]["stop"],
+            # Check sequence compatibility
+            component_root_version = component_root.attr_str("version")
+            if component_root_version not in [VERSION]:
+                raise RuntimeError(
+                    f"Can't load V4 component store {component_store_path} with incompatible data version {component_root_version}"
                 )
-                component_store_generic_meta_data = component_root_attrs["generic_meta_data"]
 
-                if not self._component_stores:
-                    self._sequence_id: str = component_store_sequence_id
-                    self._sequence_timestamp_interval_us = component_store_sequence_timestamp_interval_us
-                    self._generic_meta_data: Dict[str, types.JsonLike] = component_store_generic_meta_data
-                    self._version: str = component_root_version
+            component_store_sequence_id = component_root.attr_str("sequence_id")
+            sequence_timestamp_interval_us = component_root.attr_dict("sequence_timestamp_interval_us")
+            component_store_sequence_timestamp_interval_us = HalfClosedInterval(
+                cast(int, sequence_timestamp_interval_us["start"]),
+                cast(int, sequence_timestamp_interval_us["stop"]),
+            )
+            component_store_generic_meta_data = component_root.attr_dict("generic_meta_data")
+
+            if not self._component_stores:
+                self._sequence_id: str = component_store_sequence_id
+                self._sequence_timestamp_interval_us = component_store_sequence_timestamp_interval_us
+                self._generic_meta_data: Dict[str, types.JsonLike] = component_store_generic_meta_data
+                self._version: str = component_root_version
 
-                if not self._sequence_id == component_store_sequence_id:
-                    raise RuntimeError("Can't load component store from different sequences")
-                if not self._sequence_timestamp_interval_us == component_store_sequence_timestamp_interval_us:
-                    raise RuntimeError("Can't load component store with different sequence timestamp intervals")
-                if not self._generic_meta_data == component_store_generic_meta_data:
-                    raise RuntimeError("Can't load component store with different generic meta-data")
-                if not self._version == component_root_version:
-                    raise RuntimeError("Can't load shards from different data versions")
+            if not self._sequence_id == component_store_sequence_id:
+                raise RuntimeError("Can't load component store from different sequences")
+            if not self._sequence_timestamp_interval_us == component_store_sequence_timestamp_interval_us:
+                raise RuntimeError("Can't load component store with different sequence timestamp intervals")
+            if not self._generic_meta_data == component_store_generic_meta_data:
+                raise RuntimeError("Can't load component store with different generic meta-data")
+            if not self._version == component_root_version:
+                raise RuntimeError("Can't load shards from different data versions")
 
-                component_group_name = component_root_attrs["component_group_name"]
-                if component_group_name in self._component_stores:
-                    raise RuntimeError(f"Component group {component_group_name} loaded multiple times")
+            component_group_name = component_root.attr_str("component_group_name")
+            if component_group_name in self._component_stores:
+                raise RuntimeError(f"Component group {component_group_name} loaded multiple times")
 
-                self._component_stores[component_group_name] = (component_root, component_store_path)
+            self._component_stores[component_group_name] = (component_root, component_store_path)
 
         # Check version-compatibility
         if self._version != VERSION:
@@ -497,15 +497,8 @@ def thread_load_component_store(component_store_upath: UPath) -> Tuple[zarr.Grou
 
     def reload_resources(self) -> None:
         """Trigger a reload of each itar store - useful to re-initialize file objects in multi-process settings"""
-        component_store: Union[zarr.Group, stores.ConsolidatedCompressedMetadataStore]
-        for component_store, _ in self._component_stores.values():
-            # unwind one layer of possible consolidated metadata store
-            if isinstance(
-                compressed_consolidated_store := component_store.store, stores.ConsolidatedCompressedMetadataStore
-            ):
-                component_store = compressed_consolidated_store
-
-            if isinstance(store := component_store.store, stores.IndexedTarStore):
+        for component_root, _ in self._component_stores.values():
+            if isinstance(store := stores.group_store(component_root), stores.IndexedTarStore):
                 store.reload_resources()
 
     @property
@@ -533,20 +526,24 @@ def open_component_readers(
         ret = {}
 
         for component_root_group, _ in self._component_stores.values():
-            if (component_group := component_root_group.get(component_reader_type.get_component_name())) is None:
+            if (component_name := component_reader_type.get_component_name()) not in component_root_group:
                 continue
 
             # instantiate a reader for each of the components
-            for component_instance_name, component_group in component_group.items():
+            for component_instance_name, component_group in component_root_group.group(component_name).groups():
                 assert component_instance_name not in ret, (
                     f"Component instance {component_instance_name} encountered multiple times"
                 )
 
                 # check if the reader supports the component version
-                if not component_reader_type.supports_component_version(component_group.attrs["component_version"]):
+                if not component_reader_type.supports_component_version(component_group.attr_str("component_version")):
                     continue
 
-                ret[component_instance_name] = component_reader_type(component_instance_name, component_group)
+                # each reader caches the nodes of its component
+                cache = None if self._node_cache_size == 0 else nodes.NodeCache(self._node_cache_size)
+                ret[component_instance_name] = component_reader_type(
+                    component_instance_name, component_group.with_cache(cache)
+                )
 
         return ret
 
@@ -557,14 +554,13 @@ def get_sequence_meta(self) -> SequenceMeta:
         component_stores_info: List[SequenceMeta.ComponentStoreMeta] = []
         for component_root_group, component_store_path in self._component_stores.values():
             components: Dict[str, Dict[str, SequenceMeta.ComponentInstanceMeta]] = {}
-            for component_name, component in component_root_group.items():
+            for component_name, component in component_root_group.groups():
                 # collect component names and instances
                 component_instances: Dict[str, SequenceMeta.ComponentInstanceMeta] = {}
-                for component_instance_name, component_instance in component.items():
-                    component_instance_attrs = component_instance.attrs
+                for component_instance_name, component_instance in component.groups():
                     component_instances[component_instance_name] = SequenceMeta.ComponentInstanceMeta(
-                        version=component_instance_attrs["component_version"],
-                        generic_meta_data=component_instance_attrs["generic_meta_data"],
+                        version=component_instance.attr_str("component_version"),
+                        generic_meta_data=component_instance.attr_dict("generic_meta_data"),
                     )
 
                 components[component_name] = component_instances
@@ -612,7 +608,7 @@ def get_component_version() -> str:
         """Returns the version of the current component writer"""
         ...
 
-    def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
+    def __init__(self, component_group: nodes.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
         """Initializes a component writer targeting the given component group and sequence time interval"""
 
         self._group = component_group
@@ -652,7 +648,7 @@ def finalize(self) -> None:
 
         Called by :meth:`SequenceComponentGroupsWriter.finalize` before
         component metadata is persisted.  Override to flush buffered data
-        (e.g., create zarr datasets/groups).
+        (e.g., create arrays / groups).
         """
         pass
 
@@ -661,7 +657,7 @@ class ComponentReader(ABC):
     """Base class for V4 component readers.
 
     Subclasses must implement :meth:`get_component_name` and
-    :meth:`supports_component_version`.  The underlying zarr group is
+    :meth:`supports_component_version`.  The underlying component group is
     accessible as ``self._group``; component metadata is available via the
     :attr:`instance_name`, :attr:`component_version`, and
     :attr:`generic_meta_data` properties.
@@ -677,14 +673,16 @@ def get_component_name() -> str:
     def supports_component_version(version: str) -> bool:
         """Returns true if the component version is supported by the reader"""
 
-    def __init__(self, component_instance_name: str, component_group: zarr.Group) -> None:
+    def __init__(self, component_instance_name: str, component_group: nodes.Group) -> None:
         """Initializes a component reader for a given component instance name and group"""
         self._instance_name = component_instance_name
         self._group = component_group
 
         # Preload component meta-data and generic data group (if existing)
-        self._component_meta_data: Dict = dict(self._group.attrs)
-        self._generic_data_group: Optional[zarr.Group] = self._group.get("generic_data")
+        self._component_meta_data: Dict[str, types.JsonLike] = self._group.attrs
+        self._generic_data_group: Optional[nodes.Group] = (
+            self._group.group("generic_data") if "generic_data" in self._group else None
+        )
 
     @property
     def instance_name(self) -> str:
@@ -694,12 +692,12 @@ def instance_name(self) -> str:
     @property
     def component_version(self) -> str:
         """Returns the component version of the loaded component"""
-        return self._component_meta_data["component_version"]
+        return cast(str, self._component_meta_data["component_version"])
 
     @property
     def generic_meta_data(self) -> Dict[str, types.JsonLike]:
         """Returns the generic meta data of the loaded component"""
-        return self._component_meta_data["generic_meta_data"]
+        return cast(Dict[str, types.JsonLike], self._component_meta_data["generic_meta_data"])
 
     def has_generic_data(self, name: str) -> bool:
         """Returns True if a named generic data array exists on this component"""
@@ -707,17 +705,24 @@ def has_generic_data(self, name: str) -> bool:
 
     def get_generic_data_names(self) -> List[str]:
         """Returns the list of all generic data array names on this component"""
-        return list(self._generic_data_group.keys()) if self._generic_data_group is not None else []
+        return self._generic_data_group.members() if self._generic_data_group is not None else []
 
     def get_generic_data(self, name: str) -> np.ndarray:
         """Returns a named generic data array from this component.
 
         Raises KeyError if the name does not exist.
         """
+        return self._generic_data_array(name).read()
+
+    async def get_generic_data_async(self, name: str) -> np.ndarray:
+        """Asynchronous :meth:`get_generic_data`"""
+        return await self._generic_data_array(name).read_async()
+
+    def _generic_data_array(self, name: str) -> nodes.Array:
         if self._generic_data_group is None:
             raise KeyError("Component has no generic_data")
-        if (generic_data := self._generic_data_group.get(name)) is not None:
-            return np.array(generic_data)
+        if name in self._generic_data_group:
+            return self._generic_data_group.array(name)
         raise KeyError(f"Generic data '{name}' not found. Available: {self.get_generic_data_names()}")
 
 
@@ -751,7 +756,7 @@ def get_component_version() -> str:
             """Returns the version of the current component writer"""
             return "v1"
 
-        def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
+        def __init__(self, component_group: nodes.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
             """Initializes the current component writer targeting the given component group and sequence time interval"""
             super().__init__(component_group, sequence_timestamp_interval_us)
 
@@ -760,8 +765,8 @@ def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us:
         def finalize(self) -> None:
             """Actually store the json-encoded pose data"""
 
-            self._group.create_group("static_poses").attrs.put(self.data["static_poses"])
-            self._group.create_group("dynamic_poses").attrs.put(self.data["dynamic_poses"])
+            self._group.create_group("static_poses", attributes=self.data["static_poses"])
+            self._group.create_group("dynamic_poses", attributes=self.data["dynamic_poses"])
 
         def store_static_pose(
             self,
@@ -859,10 +864,18 @@ def supports_component_version(version: str) -> bool:
             """Returns true if the component version is supported by the reader"""
             return version == "v1"
 
+        def _static_poses(self) -> Dict[str, Dict[str, Any]]:
+            """Returns the json-encoded static poses data"""
+            return cast(Dict[str, Dict[str, Any]], self._group.group("static_poses").attrs)
+
+        def _dynamic_poses(self) -> Dict[str, Dict[str, Any]]:
+            """Returns the json-encoded dynamic poses data"""
+            return cast(Dict[str, Dict[str, Any]], self._group.group("dynamic_poses").attrs)
+
         def get_static_poses(self) -> Generator[Tuple[Tuple[str, str], npt.NDArray[np.floating]]]:
             """Returns all static poses (rigid transformations) between named coordinate frames, if available"""
 
-            for key, static_pose in self._group["static_poses"].attrs.items():
+            for key, static_pose in self._static_poses().items():
                 yield ast.literal_eval(key), np.array(static_pose["pose"], dtype=static_pose["dtype"])
 
         def get_dynamic_poses(
@@ -870,7 +883,7 @@ def get_dynamic_poses(
         ) -> Generator[Tuple[Tuple[str, str], Tuple[npt.NDArray[np.floating], npt.NDArray[np.uint64]]]]:
             """Returns all dynamic poses (time-dependent rigid transformations) between named coordinate frames, if available"""
 
-            for key, dynamic_poses in self._group["dynamic_poses"].attrs.items():
+            for key, dynamic_poses in self._dynamic_poses().items():
                 yield (
                     ast.literal_eval(key),
                     (
@@ -883,7 +896,7 @@ def get_static_pose(self, source_frame_id: str, target_frame_id: str) -> npt.NDA
             """Returns static pose (rigid transformation) between two named coordinate frames, if available"""
 
             if (
-                static_pose := self._group["static_poses"].attrs.get(
+                static_pose := self._static_poses().get(
                     key := str((validate_frame_name(source_frame_id), validate_frame_name(target_frame_id)))
                 )
             ) is None:
@@ -897,7 +910,7 @@ def get_dynamic_pose(
             """Returns dynamic poses (time-dependent rigid transformations) between two named coordinate frames, if available"""
 
             if (
-                dynamic_poses := self._group["dynamic_poses"].attrs.get(
+                dynamic_poses := self._dynamic_poses().get(
                     key := str((validate_frame_name(source_frame_id), validate_frame_name(target_frame_id)))
                 )
             ) is None:
@@ -926,7 +939,7 @@ def get_component_version() -> str:
             """Returns the version of the current intrinsic calibration component"""
             return "v1"
 
-        def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
+        def __init__(self, component_group: nodes.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
             """Initializes the current component writer targeting the given component group and sequence time interval"""
             super().__init__(component_group, sequence_timestamp_interval_us)
 
@@ -945,7 +958,7 @@ def store_camera_intrinsics(
 
             meta_data = types.encode_camera_model_parameters(camera_model_parameters)
 
-            self._cameras_group.create_group(camera_id).attrs.put(meta_data)
+            self._cameras_group.create_group(camera_id, attributes=meta_data)
 
             return self
 
@@ -960,7 +973,7 @@ def store_lidar_intrinsics(
             # Prepare meta-data containing the serialization of the mandatory lidar model
             meta_data = types.encode_lidar_model_parameters(lidar_model_parameters)
 
-            self._lidars_group.create_group(lidar_id).attrs.put(meta_data)
+            self._lidars_group.create_group(lidar_id, attributes=meta_data)
 
             return self
 
@@ -979,16 +992,14 @@ def supports_component_version(version: str) -> bool:
 
         def get_camera_model_parameters(self, camera_id: str) -> types.CameraModelParameters:
             """Returns the camera model associated with the requested camera sensor"""
-            return types.decode_camera_model_parameters(cast(zarr.Group, self._group["cameras"][camera_id]).attrs)
+            return types.decode_camera_model_parameters(self._group.group(f"cameras/{camera_id}").attrs)
 
         def get_lidar_model_parameters(self, lidar_id: str) -> Optional[types.LidarModelParameters]:
             """Returns the lidar model associated with the requested lidar sensor"""
-            lidars_group = self._group["lidars"]
-
-            if lidar_id not in lidars_group:
+            if lidar_id not in self._group.group("lidars"):
                 return None
 
-            return types.decode_lidar_model_parameters(cast(zarr.Group, lidars_group[lidar_id]).attrs)
+            return types.decode_lidar_model_parameters(self._group.group(f"lidars/{lidar_id}").attrs)
 
 
 class MasksComponent:
@@ -1009,7 +1020,7 @@ def get_component_version() -> str:
             """Returns the version of the current sensor masks component"""
             return "v1"
 
-        def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
+        def __init__(self, component_group: nodes.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
             """Initializes the current component writer targeting the given component group and sequence time interval"""
             super().__init__(component_group, sequence_timestamp_interval_us)
 
@@ -1024,8 +1035,8 @@ def store_camera_masks(
             """Store camera-associated masks"""
 
             # Store mask names
-            (camera_grp := self._cameras_group.create_group(camera_id)).attrs.put(
-                {"mask_names": list(mask_images.keys())}
+            camera_grp = self._cameras_group.create_group(
+                camera_id, attributes={"mask_names": list(mask_images.keys())}
             )
 
             # Store mask images
@@ -1034,9 +1045,7 @@ def store_camera_masks(
                     FORMAT = "png"
                     mask_image.save(buffer, format=FORMAT, optimize=True)  # encodes as png
                     # store mask data (uncompressed, as already encoded)
-                    camera_grp.create_dataset(mask_name, data=np.asarray(buffer.getvalue()), compressor=None).attrs[
-                        "format"
-                    ] = FORMAT
+                    camera_grp.create_bytes_array(mask_name, buffer.getvalue(), attributes={"format": FORMAT})
 
             return self
 
@@ -1056,14 +1065,23 @@ def supports_component_version(version: str) -> bool:
         def get_camera_mask_names(self, camera_id: str) -> List[str]:
             """Returns all constant camera mask names"""
 
-            return list(cast(zarr.Group, self._group["cameras"][camera_id]).attrs.get("mask_names", []))
+            return list(cast(List[str], self._group.group(f"cameras/{camera_id}").attrs.get("mask_names", [])))
 
         def get_camera_mask_image(self, camera_id: str, mask_name: str) -> PILImage.Image:
             """Returns constant named camera mask image"""
 
-            mask_dataset = cast(zarr.Array, cast(zarr.Group, self._group["cameras"][camera_id])[mask_name])
+            mask_array = self._group.array(f"cameras/{camera_id}/{mask_name}")
 
-            return PILImage.open(io.BytesIO(cast(np.bytes_, mask_dataset[()])), formats=[mask_dataset.attrs["format"]])
+            return PILImage.open(io.BytesIO(mask_array.read_bytes()), formats=[mask_array.attr_str("format")])
+
+        async def get_camera_mask_image_async(self, camera_id: str, mask_name: str) -> PILImage.Image:
+            """Asynchronous :meth:`get_camera_mask_image`"""
+
+            mask_array = self._group.array(f"cameras/{camera_id}/{mask_name}")
+
+            return PILImage.open(
+                io.BytesIO(await mask_array.read_bytes_async()), formats=[mask_array.attr_str("format")]
+            )
 
         def get_camera_mask_images(self, camera_id: str) -> Generator[Tuple[str, PILImage.Image]]:
             """Returns all constant named camera mask images"""
@@ -1075,7 +1093,7 @@ def get_camera_mask_images(self, camera_id: str) -> Generator[Tuple[str, PILImag
 class BaseSensorComponentWriter(ComponentWriter):
     """Base class for all sensor component writers"""
 
-    def __init__(self, component_group: zarr.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
+    def __init__(self, component_group: nodes.Group, sequence_timestamp_interval_us: HalfClosedInterval) -> None:
         """Initializes the current component writer targeting the given component group and sequence time interval"""
         super().__init__(component_group, sequence_timestamp_interval_us)
 
@@ -1103,21 +1121,16 @@ def finalize(self) -> None:
         )
 
         # Store as meta-data of frames group
-        self._frames_group.attrs.put({"frames_timestamps_us": frames_timestamps_us.tolist()})
+        self._frames_group.update_attrs({"frames_timestamps_us": frames_timestamps_us.tolist()})
 
     def _get_frame_group(
         self,
         # end-of-frame timestamp, or start-of-frame / end-of-frame timestamps
         timestamps_us: Union[int, npt.NDArray[np.uint64]],
-    ) -> zarr.Group:
+    ) -> nodes.Group:
         """Returns the group of a frame, initializing it if required"""
 
-        if isinstance(timestamps_us, np.ndarray):
-            frame_id = cast(np.ndarray, timestamps_us)[1].item()  # end-of-frame timestamp is frame ID
-        else:
-            frame_id = timestamps_us
-
-        return self._frames_group.require_group(str(frame_id))
+        return self._frames_group.require_group(str(_frame_id(timestamps_us)))
 
     def _store_base_frame(
         self,
@@ -1126,7 +1139,7 @@ def _store_base_frame(
         # generic per-frame data (key-value pairs, *not* dimension / dtype validated) and meta-data
         generic_data: Dict[str, npt.NDArray[Any]],
         generic_meta_data: Dict[str, types.JsonLike],
-    ) -> zarr.Group:
+    ) -> nodes.Group:
         # Sanity / timestamp consistency checks
         assert np.shape(frame_timestamps_us) == (2,)
         assert frame_timestamps_us.dtype == np.dtype("uint64")
@@ -1149,16 +1162,11 @@ def _store_base_frame(
         self._frames_timestamps_us[frame_timestamps_us[1].item()] = frame_timestamps_us[0].item()
 
         # Store additional generic frame data and meta-data (not dimension / dtype checked)
-        (frame_generic_data_group := frame_group.create_group("generic_data")).attrs.put(generic_meta_data)
+        frame_generic_data_group = frame_group.create_group("generic_data", attributes=generic_meta_data)
         for name, value in generic_data.items():
-            frame_generic_data_group.create_dataset(
-                name,
-                data=value,
-                # we are not accessing sub-ranges, so disable chunking
-                chunks=_normalize_chunks(value.shape),
-                # use compression that is fast to decode on modern hardware
-                compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-            )
+            # we are not accessing sub-ranges, so store a single chunk (default) - use the default compression,
+            # which is fast to decode on modern hardware
+            frame_generic_data_group.create_array(name, value)
 
         return frame_group
 
@@ -1166,7 +1174,7 @@ def _store_base_frame(
 class BaseSensorComponentReader(ComponentReader):
     """Base class for all sensor component readers"""
 
-    def __init__(self, component_instance_name: str, component_group: zarr.Group) -> None:
+    def __init__(self, component_instance_name: str, component_group: nodes.Group) -> None:
         """Initializes a component reader for a given component instance name and group"""
         super().__init__(component_instance_name, component_group)
 
@@ -1174,7 +1182,9 @@ def __init__(self, component_instance_name: str, component_group: zarr.Group) ->
             raise RuntimeError("Sensor component doesn't contain any frames")
 
         # preload frame timestamps and create map
-        self._frames_timestamps_us = np.array(self._group["frames"].attrs["frames_timestamps_us"], dtype=np.uint64)
+        self._frames_timestamps_us = np.array(
+            self._group.group("frames").attr_list("frames_timestamps_us"), dtype=np.uint64
+        )
         self._frame_end_to_frame_timestamps_us = {
             end: np.array([self._frames_timestamps_us[i, 0], end], dtype=np.uint64)
             for i, end in enumerate(self._frames_timestamps_us[:, 1])
@@ -1184,19 +1194,32 @@ def _get_frame_group(
         self,
         # end-of-frame timestamp, or start-of-frame / end-of-frame timestamps
         timestamps_us: Union[int, npt.NDArray[np.uint64]],
-    ) -> zarr.Group:
+    ) -> nodes.Group:
         """Returns the group of a frame"""
 
-        if isinstance(timestamps_us, np.ndarray):
-            frame_id = cast(np.ndarray, timestamps_us)[1].item()  # end-of-frame timestamp is frame ID
-        else:
-            frame_id = timestamps_us
+        return self._group.group(self._frame_path(timestamps_us))
+
+    @staticmethod
+    def _frame_path(
+        # end-of-frame timestamp, or start-of-frame / end-of-frame timestamps
+        timestamps_us: Union[int, npt.NDArray[np.uint64]],
+        # optional name of a frame member
+        name: Optional[str] = None,
+    ) -> str:
+        """Returns the path of a frame group (or of one of its members) within the component group"""
+
+        frame_path = f"frames/{_frame_id(timestamps_us)}"
 
-        return cast(zarr.Group, self._group["frames"][str(frame_id)])
+        return frame_path if name is None else f"{frame_path}/{name}"
+
+    def _get_frame_generic_data_group(self, timestamp_us: int) -> nodes.Group:
+        """Returns the generic data group of a frame"""
+
+        return self._group.group(self._frame_path(timestamp_us, "generic_data"))
 
     @property
     def frames_timestamps_us(self) -> npt.NDArray[np.uint64]:
-        return np.array(self._group["frames"].attrs["frames_timestamps_us"], dtype=np.uint64)
+        return np.array(self._group.group("frames").attr_list("frames_timestamps_us"), dtype=np.uint64)
 
     @property
     def frames_count(self) -> int:
@@ -1209,7 +1232,7 @@ def get_frame_timestamps_us(self, timestamp_us: int) -> npt.NDArray[np.uint64]:
     def get_frame_generic_data_names(self, timestamp_us: int) -> List[str]:
         """List of all generic frame-data names"""
 
-        return list(cast(zarr.Group, self._get_frame_group(timestamp_us)["generic_data"]).keys())
+        return self._get_frame_generic_data_group(timestamp_us).members()
 
     def has_frame_generic_data(self, timestamp_us: int, name: str) -> bool:
         """Signals if named generic frame-data exists"""
@@ -1219,12 +1242,17 @@ def has_frame_generic_data(self, timestamp_us: int, name: str) -> bool:
     def get_frame_generic_data(self, timestamp_us: int, name: str) -> npt.NDArray[Any]:
         """Returns generic frame-data for a specific frame and name"""
 
-        return np.array(self._get_frame_group(timestamp_us)["generic_data"][name])
+        return self._get_frame_generic_data_group(timestamp_us).array(name).read()
+
+    async def get_frame_generic_data_async(self, timestamp_us: int, name: str) -> npt.NDArray[Any]:
+        """Asynchronous :meth:`get_frame_generic_data`"""
+
+        return await self._get_frame_generic_data_group(timestamp_us).array(name).read_async()
 
     def get_frame_generic_meta_data(self, timestamp_us: int) -> Dict[str, types.JsonLike]:
         """Returns generic frame meta-data for a specific frame"""
 
-        return dict(self._get_frame_group(timestamp_us)["generic_data"].attrs)
+        return self._get_frame_generic_data_group(timestamp_us).attrs
 
 
 class BaseRayBundleSensorComponentWriter(BaseSensorComponentWriter):
@@ -1244,21 +1272,16 @@ def _store_frame_ray_bundle(
     ) -> None:
         ## Initialize ray bundle group
         frame_group = self._get_frame_group(frame_timestamps_us)
-        (ray_bundle_group := frame_group.create_group("ray_bundle")).attrs.put({"n_rays": n_rays})
+        ray_bundle_group = frame_group.create_group("ray_bundle", attributes={"n_rays": n_rays})
 
-        # Store per-ray data
+        # Store per-ray data (zero-sized chunk dimensions are clamped to 1, default compression is fast to decode on
+        # modern hardware)
         for name, (ray_data_data, chunks) in ray_data.items():
             assert len(ray_data_data) == n_rays, f"{name} doesn't have required ray count"
-            ray_bundle_group.create_dataset(
-                name,
-                data=ray_data_data,
-                chunks=_normalize_chunks(chunks),
-                # use compression that is fast to decode on modern hardware
-                compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-            )
+            ray_bundle_group.create_array(name, ray_data_data, chunks=chunks)
 
         ## Initialize ray bundle returns group
-        (ray_bundle_returns_group := frame_group.create_group("ray_bundle_returns")).attrs.put({"n_returns": n_returns})
+        ray_bundle_returns_group = frame_group.create_group("ray_bundle_returns", attributes={"n_returns": n_returns})
 
         # Store per-return data
         absent_mask = None
@@ -1291,13 +1314,7 @@ def _store_frame_ray_bundle(
                 # validate absent mask consistency
                 assert np.array_equal(absent_mask, local_absent_mask), f"Inconsistent NaN masks in return data {name}"
 
-            ray_bundle_returns_group.create_dataset(
-                name,
-                data=return_data_data,
-                chunks=_normalize_chunks(chunks),
-                # use compression that is fast to decode on modern hardware
-                compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-            )
+            ray_bundle_returns_group.create_array(name, return_data_data, chunks=chunks)
 
         if absent_mask is None:
             # Initialize empty absent mask if no per-return data was provided
@@ -1305,34 +1322,32 @@ def _store_frame_ray_bundle(
 
         valid_mask_packed = np.packbits(~absent_mask)
 
-        frame_group.create_dataset(
+        # we are not accessing sub-ranges, so store a single chunk (default)
+        frame_group.create_array(
             "ray_bundle_returns_valid_mask_packed",
-            data=valid_mask_packed,
-            # we are not accessing sub-ranges, so disable chunking
-            chunks=_normalize_chunks(valid_mask_packed.shape),
-            # use compression that is fast to decode on modern hardware
-            compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-        ).attrs.put({"n_returns": n_returns, "n_rays": n_rays})
+            valid_mask_packed,
+            attributes={"n_returns": n_returns, "n_rays": n_rays},
+        )
 
 
 class BaseRayBundleSensorComponentReader(BaseSensorComponentReader):
     """Base class for all ray bundle sensor component readers"""
 
     # per-ray data
-    def _get_ray_bundle_group(self, timestamp_us: int) -> zarr.Group:
+    def _get_ray_bundle_group(self, timestamp_us: int) -> nodes.Group:
         """Returns the ray bundle group of a frame"""
 
-        return cast(zarr.Group, self._get_frame_group(timestamp_us)["ray_bundle"])
+        return self._group.group(self._frame_path(timestamp_us, "ray_bundle"))
 
     def get_frame_ray_bundle_count(self, timestamp_us: int) -> int:
         """Returns the number of rays and ray returns for a specific frame"""
 
-        return self._get_ray_bundle_group(timestamp_us).attrs["n_rays"]
+        return self._get_ray_bundle_group(timestamp_us).attr_int("n_rays")
 
     def get_frame_ray_bundle_data_names(self, timestamp_us: int) -> List[str]:
         """List of all ray bundle data names for a frame"""
 
-        return list(self._get_ray_bundle_group(timestamp_us).keys())
+        return self._get_ray_bundle_group(timestamp_us).members()
 
     def has_frame_ray_bundle_data(self, timestamp_us: int, name: str) -> bool:
         """Signals if named ray bundle data exists for a frame"""
@@ -1342,23 +1357,28 @@ def has_frame_ray_bundle_data(self, timestamp_us: int, name: str) -> bool:
     def get_frame_ray_bundle_data(self, timestamp_us: int, name: str) -> npt.NDArray[Any]:
         """Returns named ray bundle data for a frame"""
 
-        return np.array(self._get_ray_bundle_group(timestamp_us)[name])
+        return self._get_ray_bundle_group(timestamp_us).array(name).read()
+
+    async def get_frame_ray_bundle_data_async(self, timestamp_us: int, name: str) -> npt.NDArray[Any]:
+        """Asynchronous :meth:`get_frame_ray_bundle_data`"""
+
+        return await self._get_ray_bundle_group(timestamp_us).array(name).read_async()
 
     # per-ray return data
-    def _get_ray_bundle_returns_group(self, timestamp_us: int) -> zarr.Group:
+    def _get_ray_bundle_returns_group(self, timestamp_us: int) -> nodes.Group:
         """Returns the ray bundle returns group of a frame"""
 
-        return cast(zarr.Group, self._get_frame_group(timestamp_us)["ray_bundle_returns"])
+        return self._group.group(self._frame_path(timestamp_us, "ray_bundle_returns"))
 
     def get_frame_ray_bundle_return_count(self, timestamp_us: int) -> int:
         """Returns the number of ray returns for a specific frame"""
 
-        return self._get_ray_bundle_returns_group(timestamp_us).attrs["n_returns"]
+        return self._get_ray_bundle_returns_group(timestamp_us).attr_int("n_returns")
 
     def get_frame_ray_bundle_return_data_names(self, timestamp_us: int) -> List[str]:
         """List of all ray bundle return data names for a frame"""
 
-        return list(self._get_ray_bundle_returns_group(timestamp_us).keys())
+        return self._get_ray_bundle_returns_group(timestamp_us).members()
 
     def has_frame_ray_bundle_return_data(self, timestamp_us: int, name: str) -> bool:
         """Signals if named ray bundle return data exists for a frame"""
@@ -1368,30 +1388,45 @@ def has_frame_ray_bundle_return_data(self, timestamp_us: int, name: str) -> bool
     def get_frame_ray_bundle_return_valid_mask(self, timestamp_us: int) -> npt.NDArray[np.bool_]:
         """Returns the per-ray return valid mask for a frame"""
 
-        valid_mask_packed = self._get_frame_group(timestamp_us)["ray_bundle_returns_valid_mask_packed"]
+        valid_mask_packed = self._valid_mask_packed_array(timestamp_us)
+        return self._unpack_valid_mask(valid_mask_packed, valid_mask_packed.read())
 
-        attrs = valid_mask_packed.attrs
-        n_returns, n_rays = attrs["n_returns"], attrs["n_rays"]
+    async def get_frame_ray_bundle_return_valid_mask_async(self, timestamp_us: int) -> npt.NDArray[np.bool_]:
+        """Asynchronous :meth:`get_frame_ray_bundle_return_valid_mask`"""
 
-        return (
-            np.unpackbits(np.array(valid_mask_packed), count=n_returns * n_rays)
-            .astype(np.bool_)
-            .reshape((n_returns, n_rays))
-        )
+        valid_mask_packed = self._valid_mask_packed_array(timestamp_us)
+        return self._unpack_valid_mask(valid_mask_packed, await valid_mask_packed.read_async())
+
+    def _valid_mask_packed_array(self, timestamp_us: int) -> nodes.Array:
+        return self._group.array(self._frame_path(timestamp_us, "ray_bundle_returns_valid_mask_packed"))
+
+    @staticmethod
+    def _unpack_valid_mask(valid_mask_packed: nodes.Array, packed: npt.NDArray[np.uint8]) -> npt.NDArray[np.bool_]:
+        n_returns, n_rays = valid_mask_packed.attr_int("n_returns"), valid_mask_packed.attr_int("n_rays")
+        return np.unpackbits(packed, count=n_returns * n_rays).astype(np.bool_).reshape((n_returns, n_rays))
 
     def get_frame_ray_bundle_return_data(
         self, timestamp_us: int, name: str, return_index: Optional[int]
     ) -> npt.NDArray[np.float32]:
         """Returns named ray bundle return data for a frame, optionally indexed by return index to accelerate data-retrieval"""
 
-        return_array = self._get_ray_bundle_returns_group(timestamp_us)[
-            name
-        ]  # only references the underlying Array, don't load it's data yet
+        # only references the underlying array, don't load its data yet
+        return_array = self._get_ray_bundle_returns_group(timestamp_us).array(name)
 
         if return_index is None:
-            return np.array(return_array[slice(return_array.shape[0]), ...])  # load all returns
+            return return_array.read()  # load all returns
         else:
-            return np.array(return_array[return_index, ...])  # load specific return only
+            return return_array.read((return_index, Ellipsis))  # load specific return only
+
+    async def get_frame_ray_bundle_return_data_async(
+        self, timestamp_us: int, name: str, return_index: Optional[int]
+    ) -> npt.NDArray[np.float32]:
+        """Asynchronous :meth:`get_frame_ray_bundle_return_data`"""
+
+        return_array = self._get_ray_bundle_returns_group(timestamp_us).array(name)
+        if return_index is None:
+            return await return_array.read_async()
+        return await return_array.read_async((return_index, Ellipsis))
 
 
 class CameraSensorComponent:
@@ -1427,9 +1462,7 @@ def store_frame(
             frame_group = self._store_base_frame(frame_timestamps_us, generic_data, generic_meta_data)
 
             # Store image data (uncompressed, as already encoded)
-            frame_group.create_dataset("image", data=np.asarray(image_binary_data), compressor=None).attrs["format"] = (
-                image_format
-            )
+            frame_group.create_bytes_array("image", image_binary_data, attributes={"format": image_format})
 
             return self
 
@@ -1449,23 +1482,31 @@ def supports_component_version(version: str) -> bool:
         class EncodedImageDataHandle:
             """References encoded image data without loading it"""
 
-            def __init__(self, image_dataset: zarr.Array) -> None:
-                self._image_dataset = image_dataset
+            def __init__(self, image_array: nodes.Array) -> None:
+                self._image_array = image_array
 
             def get_data(self) -> types.EncodedImageData:
                 """Loads the referenced encoded image data to memory"""
+                return types.EncodedImageData(self._image_array.read_bytes(), self._image_array.attr_str("format"))
+
+            async def get_data_async(self) -> types.EncodedImageData:
+                """Asynchronous :meth:`get_data`"""
                 return types.EncodedImageData(
-                    cast(np.bytes_, self._image_dataset[()]), self._image_dataset.attrs["format"]
+                    await self._image_array.read_bytes_async(), self._image_array.attr_str("format")
                 )
 
         def get_frame_handle(self, timestamp_us: int) -> EncodedImageDataHandle:
             """Returns the frame's encoded image data"""
-            return self.EncodedImageDataHandle(cast(zarr.Array, self._get_frame_group(timestamp_us)["image"]))
+            return self.EncodedImageDataHandle(self._group.array(self._frame_path(timestamp_us, "image")))
 
         def get_frame_data(self, timestamp_us: int) -> types.EncodedImageData:
             """Returns the frame's encoded image data"""
             return self.get_frame_handle(timestamp_us).get_data()
 
+        async def get_frame_data_async(self, timestamp_us: int) -> types.EncodedImageData:
+            """Asynchronous :meth:`get_frame_data`"""
+            return await self.get_frame_handle(timestamp_us).get_data_async()
+
         def get_frame_image(self, timestamp_us: int) -> PILImage.Image:
             """Returns the frame's decoded image data"""
             return self.get_frame_data(timestamp_us).get_decoded_image()
@@ -1706,7 +1747,7 @@ def store_observations(
                 )
                 obs_dict_list.append(obs.to_dict())
 
-            self._group.create_group("cuboids").attrs.put({"cuboid_track_observations": obs_dict_list})
+            self._group.create_group("cuboids", attributes={"cuboid_track_observations": obs_dict_list})
 
             return self
 
@@ -1726,8 +1767,8 @@ def supports_component_version(version: str) -> bool:
         def get_observations(self) -> Generator[types.CuboidTrackObservation]:
             """Returns all stored cuboid track observations"""
 
-            for obs in self._group["cuboids"].attrs["cuboid_track_observations"]:
-                yield types.CuboidTrackObservation.from_dict(obs)
+            for obs in self._group.group("cuboids").attr_list("cuboid_track_observations"):
+                yield types.CuboidTrackObservation.from_dict(cast(Dict[str, Any], obs))
 
 
 class PointCloudsComponent:
@@ -1771,7 +1812,7 @@ def get_component_version() -> str:
 
         def __init__(
             self,
-            component_group: zarr.Group,
+            component_group: nodes.Group,
             sequence_timestamp_interval_us: HalfClosedInterval,
             coordinate_unit: PointCloud.CoordinateUnit,
             attribute_schemas: Optional[Dict[str, PointCloudsComponent.AttributeSchema]] = None,
@@ -1785,7 +1826,7 @@ def __init__(
 
             # Pre-create the pcs group
             self._pcs_group = self._group.require_group("pcs")
-            self._pcs_group.attrs.put(
+            self._pcs_group.update_attrs(
                 {
                     "coordinate_unit": coordinate_unit.name,
                     "attribute_schemas": {name: s.to_dict() for name, s in self._attribute_schemas.items()},
@@ -1812,7 +1853,6 @@ def store_pc(
             attributes = unpack_optional(attributes, default={})
             generic_data = unpack_optional(generic_data, default={})
             generic_meta_data = unpack_optional(generic_meta_data, default={})
-            compressor = Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE)
 
             # -- Validate xyz --
             assert xyz.dtype == np.dtype("float32")
@@ -1843,7 +1883,7 @@ def store_pc(
             pc_group = self._pcs_group.require_group(str(len(self._pc_timestamps)))
 
             # Store per-pc metadata (timestamp lives in source-level pc_timestamps_us array)
-            pc_group.attrs.put(
+            pc_group.update_attrs(
                 {
                     "reference_frame_id": reference_frame_id,
                     "generic_meta_data": generic_meta_data,
@@ -1851,43 +1891,23 @@ def store_pc(
             )
 
             # Store xyz
-            pc_group.create_dataset(
-                "xyz",
-                data=xyz,
-                chunks=_normalize_chunks(xyz.shape, require_nonzero_dims=(1,)),
-                compressor=compressor,
-            )
+            _check_nonzero_dims(xyz.shape, (1,), "xyz")
+            pc_group.create_array("xyz", xyz)
 
             # Store schema-declared attributes
             for attr_name, attr_array in attributes.items():
-                pc_group.create_dataset(
-                    attr_name,
-                    data=attr_array,
-                    chunks=_normalize_chunks(attr_array.shape),
-                    compressor=compressor,
-                )
+                pc_group.create_array(attr_name, attr_array)
 
             # Store generic data
             gd_group = pc_group.require_group("generic_data")
             for gd_name, gd_array in generic_data.items():
-                gd_group.create_dataset(
-                    gd_name,
-                    data=gd_array,
-                    chunks=_normalize_chunks(gd_array.shape),
-                    compressor=compressor,
-                )
+                gd_group.create_array(gd_name, gd_array)
 
             self._pc_timestamps.append(reference_frame_timestamp_us)
 
         def finalize(self) -> None:
             """Write pc_timestamps_us array (derived from per-pc reference_frame_timestamp_us values)."""
-            ts_array = np.array(self._pc_timestamps, dtype=np.uint64)
-            self._group.create_dataset(
-                "pc_timestamps_us",
-                data=ts_array,
-                chunks=_normalize_chunks(ts_array.shape),
-                compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-            )
+            self._group.create_array("pc_timestamps_us", np.array(self._pc_timestamps, dtype=np.uint64))
 
     # --------------------------------------------------------------------------
     # Reader
@@ -1904,16 +1924,16 @@ def get_component_name() -> str:
         def supports_component_version(version: str) -> bool:
             return version == "v1"
 
-        def __init__(self, component_instance_name: str, component_group: zarr.Group) -> None:
+        def __init__(self, component_instance_name: str, component_group: nodes.Group) -> None:
             super().__init__(component_instance_name, component_group)
 
-            self._pc_timestamps_us: npt.NDArray[np.uint64] = np.array(self._group["pc_timestamps_us"][:])
+            self._pc_timestamps_us: npt.NDArray[np.uint64] = self._group.array("pc_timestamps_us").read()
 
-            pcs_attrs = self._group["pcs"].attrs
-            self._coordinate_unit = PointCloud.CoordinateUnit[pcs_attrs["coordinate_unit"]]
+            pcs_group = self._group.group("pcs")
+            self._coordinate_unit = PointCloud.CoordinateUnit[pcs_group.attr_str("coordinate_unit")]
             self._attribute_schemas: Dict[str, PointCloudsComponent.AttributeSchema] = {
-                name: PointCloudsComponent.AttributeSchema.from_dict(s)
-                for name, s in pcs_attrs["attribute_schemas"].items()
+                name: PointCloudsComponent.AttributeSchema.from_dict(cast(Dict[str, Any], s))
+                for name, s in pcs_group.attr_dict("attribute_schemas").items()
             }
 
         # -- properties --------------------------------------------------------
@@ -1942,32 +1962,52 @@ def get_attribute_schema(self, name: str) -> PointCloudsComponent.AttributeSchem
 
         # -- per-pc data access ------------------------------------------------
 
-        def _pc_group(self, pc_index: int) -> zarr.Group:
-            return cast(zarr.Group, self._group["pcs"][str(pc_index)])
+        @staticmethod
+        def _pc_path(pc_index: int, name: Optional[str] = None) -> str:
+            """Returns the path of a point cloud group (or of one of its members) within the component group"""
+            return f"pcs/{pc_index}" if name is None else f"pcs/{pc_index}/{name}"
+
+        def _pc_group(self, pc_index: int, name: Optional[str] = None) -> nodes.Group:
+            return self._group.group(self._pc_path(pc_index, name))
+
+        def _pc_array(self, pc_index: int, name: str) -> nodes.Array:
+            return self._group.array(self._pc_path(pc_index, name))
 
         def get_pc_xyz(self, pc_index: int) -> npt.NDArray[np.float32]:
-            return np.array(self._pc_group(pc_index)["xyz"][:])
+            return self._pc_array(pc_index, "xyz").read()
+
+        async def get_pc_xyz_async(self, pc_index: int) -> npt.NDArray[np.float32]:
+            """Asynchronous :meth:`get_pc_xyz`"""
+            return await self._pc_array(pc_index, "xyz").read_async()
 
         def get_pc_attribute(self, pc_index: int, name: str) -> npt.NDArray[Any]:
-            return np.array(self._pc_group(pc_index)[name][:])
+            return self._pc_array(pc_index, name).read()
+
+        async def get_pc_attribute_async(self, pc_index: int, name: str) -> npt.NDArray[Any]:
+            """Asynchronous :meth:`get_pc_attribute`"""
+            return await self._pc_array(pc_index, name).read_async()
 
         def get_pc_reference_frame_id(self, pc_index: int) -> str:
-            return str(self._pc_group(pc_index).attrs["reference_frame_id"])
+            return self._pc_group(pc_index).attr_str("reference_frame_id")
 
         def get_pc_reference_frame_timestamp_us(self, pc_index: int) -> int:
-            return int(self._group["pc_timestamps_us"][pc_index])
+            return int(self._group.array("pc_timestamps_us").read(pc_index))
 
         def get_pc_generic_data_names(self, pc_index: int) -> List[str]:
-            return list(cast(zarr.Group, self._pc_group(pc_index)["generic_data"]).keys())
+            return self._pc_group(pc_index, "generic_data").members()
 
         def has_pc_generic_data(self, pc_index: int, name: str) -> bool:
-            return name in self._pc_group(pc_index)["generic_data"]
+            return name in self._pc_group(pc_index, "generic_data")
 
         def get_pc_generic_data(self, pc_index: int, name: str) -> npt.NDArray[Any]:
-            return np.array(self._pc_group(pc_index)["generic_data"][name][:])
+            return self._pc_array(pc_index, f"generic_data/{name}").read()
+
+        async def get_pc_generic_data_async(self, pc_index: int, name: str) -> npt.NDArray[Any]:
+            """Asynchronous :meth:`get_pc_generic_data`"""
+            return await self._pc_array(pc_index, f"generic_data/{name}").read_async()
 
         def get_pc_generic_meta_data(self, pc_index: int) -> Dict[str, types.JsonLike]:
-            return dict(self._pc_group(pc_index).attrs.get("generic_meta_data", {}))
+            return cast(Dict[str, types.JsonLike], self._pc_group(pc_index).attrs.get("generic_meta_data", {}))
 
 
 class CameraLabelsComponent:
@@ -1992,7 +2032,7 @@ def get_component_version() -> str:
 
         def __init__(
             self,
-            component_group: zarr.Group,
+            component_group: nodes.Group,
             sequence_timestamp_interval_us: HalfClosedInterval,
             descriptor: types.CameraLabelDescriptor,
         ) -> None:
@@ -2002,7 +2042,7 @@ def __init__(
 
             # Initialize labels group and timestamps list
             self._labels_group = self._group.require_group("labels")
-            self._labels_group.attrs.put(
+            self._labels_group.update_attrs(
                 {
                     "descriptor": descriptor.to_dict(),
                 }
@@ -2030,7 +2070,6 @@ def store_label(
                 Optional per-label metadata.
             """
             generic_meta_data = unpack_optional(generic_meta_data, default={})
-            compressor = Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE)
 
             # Sanity checks
             assert timestamp_us in self._sequence_timestamp_interval_us, (
@@ -2067,38 +2106,32 @@ def store_label(
                         q.quantized_dtype
                     )
 
-                label_group.create_dataset(
-                    "data",
-                    data=stored,
-                    chunks=_normalize_chunks(stored.shape, require_nonzero_dims=(0, 1)),
-                    compressor=compressor,
-                )
+                _check_nonzero_dims(stored.shape, (0, 1), "label data")
+                label_group.create_array("data", stored)
 
             elif self._descriptor.label_schema.encoding == types.LabelEncoding.IMAGE_ENCODED:
                 assert isinstance(data, bytes), "IMAGE_ENCODED encoding requires bytes"
 
-                label_group.create_dataset(
+                # store encoded image data as uncompressed 1-d uint8 array (as already encoded), with chunks
+                # determined by zarr-python's chunk shape heuristic
+                label_group.create_array(
                     "data",
-                    data=np.asarray(bytearray(data), dtype=np.uint8),
-                    compressor=None,
-                ).attrs["format"] = self._descriptor.label_schema.encoded_format
+                    np.frombuffer(data, dtype=np.uint8),
+                    chunks="auto",
+                    compression=None,
+                    attributes={"format": self._descriptor.label_schema.encoded_format},
+                )
 
             else:
                 raise ValueError(f"Unsupported label encoding: {self._descriptor.label_schema.encoding}")
 
-            label_group.attrs["generic_meta_data"] = generic_meta_data
+            label_group.update_attrs({"generic_meta_data": generic_meta_data})
 
             self._timestamps.append(timestamp_us)
 
         def finalize(self) -> None:
             """Write sorted timestamps_us array."""
-            ts_array = np.array(sorted(self._timestamps), dtype=np.uint64)
-            self._group.create_dataset(
-                "timestamps_us",
-                data=ts_array,
-                chunks=_normalize_chunks(ts_array.shape),
-                compressor=Blosc(cname="lz4", clevel=5, shuffle=Blosc.BITSHUFFLE),
-            )
+            self._group.create_array("timestamps_us", np.array(sorted(self._timestamps), dtype=np.uint64))
 
     # --------------------------------------------------------------------------
     # Reader
@@ -2115,12 +2148,14 @@ def get_component_name() -> str:
         def supports_component_version(version: str) -> bool:
             return version == "v1"
 
-        def __init__(self, component_instance_name: str, component_group: zarr.Group) -> None:
+        def __init__(self, component_instance_name: str, component_group: nodes.Group) -> None:
             super().__init__(component_instance_name, component_group)
 
-            self._timestamps_us: npt.NDArray[np.uint64] = np.array(self._group["timestamps_us"][:])
+            self._timestamps_us: npt.NDArray[np.uint64] = self._group.array("timestamps_us").read()
             self._timestamp_to_index: Dict[int, int] = {int(ts): i for i, ts in enumerate(self._timestamps_us)}
-            self._descriptor = types.CameraLabelDescriptor.from_dict(self._group["labels"].attrs["descriptor"])
+            self._descriptor = types.CameraLabelDescriptor.from_dict(
+                self._group.group("labels").attr_dict("descriptor")
+            )
 
         # -- properties --------------------------------------------------------
 
@@ -2138,11 +2173,11 @@ def timestamps_us(self) -> npt.NDArray[np.uint64]:
 
         # -- per-label access --------------------------------------------------
 
-        def _label_group(self, timestamp_us: int) -> zarr.Group:
+        def _label_group(self, timestamp_us: int) -> nodes.Group:
             assert timestamp_us in self._timestamp_to_index, (
                 f"Unknown timestamp: {timestamp_us}. Available: {list(self._timestamp_to_index.keys())[:5]}..."
             )
-            return cast(zarr.Group, self._group["labels"][str(timestamp_us)])
+            return self._group.group(f"labels/{timestamp_us}")
 
         class CameraLabelHandle:
             """References label data without eagerly loading it.
@@ -2153,7 +2188,7 @@ class CameraLabelHandle:
 
             def __init__(
                 self,
-                label_group: zarr.Group,
+                label_group: nodes.Group,
                 descriptor: types.CameraLabelDescriptor,
                 timestamp_us: int,
                 generic_meta_data: Dict[str, types.JsonLike],
@@ -2181,35 +2216,50 @@ def get_data(self) -> npt.NDArray[Any]:
                 For RAW encoding, applies de-quantization if specified in the schema.
                 For IMAGE_ENCODED encoding, decodes the image bytes via PIL.
                 """
+                data = self._label_group.array("data")
                 if self._descriptor.label_schema.encoding == types.LabelEncoding.RAW:
-                    arr = np.array(self._label_group["data"][:])
-
-                    # De-quantize if configured
-                    if (q := self._descriptor.label_schema.quantization) is not None:
-                        arr = (arr.astype(q.intermediate_dtype) * q.scale + q.offset).astype(
-                            self._descriptor.label_schema.dtype
-                        )
-
-                    return arr
-
+                    return self._decode_raw(data.read())
                 elif self._descriptor.label_schema.encoding == types.LabelEncoding.IMAGE_ENCODED:
-                    raw_bytes = bytes(self._label_group["data"][:])
-                    image = PILImage.open(io.BytesIO(raw_bytes))
-
-                    return np.asarray(image, dtype=self._descriptor.label_schema.dtype)
+                    return self._decode_image(data.read_bytes())
+                else:
+                    raise ValueError(f"Unsupported label encoding: {self._descriptor.label_schema.encoding}")
 
+            async def get_data_async(self) -> npt.NDArray[Any]:
+                """Asynchronous :meth:`get_data`"""
+                data = self._label_group.array("data")
+                if self._descriptor.label_schema.encoding == types.LabelEncoding.RAW:
+                    return self._decode_raw(await data.read_async())
+                elif self._descriptor.label_schema.encoding == types.LabelEncoding.IMAGE_ENCODED:
+                    return self._decode_image(await data.read_bytes_async())
                 else:
                     raise ValueError(f"Unsupported label encoding: {self._descriptor.label_schema.encoding}")
 
+            def _decode_raw(self, arr: npt.NDArray[Any]) -> npt.NDArray[Any]:
+                # De-quantize if configured
+                if (q := self._descriptor.label_schema.quantization) is not None:
+                    arr = (arr.astype(q.intermediate_dtype) * q.scale + q.offset).astype(
+                        self._descriptor.label_schema.dtype
+                    )
+                return arr
+
+            def _decode_image(self, raw_bytes: bytes) -> npt.NDArray[Any]:
+                return np.asarray(PILImage.open(io.BytesIO(raw_bytes)), dtype=self._descriptor.label_schema.dtype)
+
             def get_encoded_data(self) -> Optional[bytes]:
                 """Return the raw encoded bytes for IMAGE_ENCODED labels, or None for RAW."""
                 if self._descriptor.label_schema.encoding == types.LabelEncoding.IMAGE_ENCODED:
-                    return bytes(self._label_group["data"][:])
+                    return self._label_group.array("data").read_bytes()
+                return None
+
+            async def get_encoded_data_async(self) -> Optional[bytes]:
+                """Asynchronous :meth:`get_encoded_data`"""
+                if self._descriptor.label_schema.encoding == types.LabelEncoding.IMAGE_ENCODED:
+                    return await self._label_group.array("data").read_bytes_async()
                 return None
 
         def get_label(self, timestamp_us: int) -> CameraLabelHandle:
             """Return a lazy handle to the label data at the given timestamp."""
             label_group = self._label_group(timestamp_us)
             return self.CameraLabelHandle(
-                label_group, self._descriptor, timestamp_us, label_group.attrs["generic_meta_data"]
+                label_group, self._descriptor, timestamp_us, label_group.attr_dict("generic_meta_data")
             )
diff --git a/ncore/impl/data/v4/components_test.py b/ncore/impl/data/v4/components_test.py
index 5ae85ba0..9187886c 100644
--- a/ncore/impl/data/v4/components_test.py
+++ b/ncore/impl/data/v4/components_test.py
@@ -16,9 +16,11 @@
 import dataclasses
 import io
 import tempfile
+import threading
 import unittest
 
-from typing import Dict, Literal, Tuple, cast
+from typing import Any, Dict, Literal, Optional, Tuple, cast
+from unittest import mock
 
 import numpy as np
 import PIL.Image as PILImage
@@ -29,6 +31,8 @@
 
 from ncore.impl.common.transformations import HalfClosedInterval
 from ncore.impl.common.util import unpack_optional
+from ncore.impl.data import nodes, stores
+from ncore.impl.data.nodes import BloscCompression
 from ncore.impl.data.types import (
     BBox3,
     BivariateWindshieldModelParameters,
@@ -65,29 +69,39 @@
 )
 
 
-@parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
-)
+#: Store types and zarr formats of the component tests (zarr formats unsupported by the installed zarr-python version are
+#: skipped)
+STORE_TYPES_AND_ZARR_FORMATS = [
+    (store_type, zarr_format) for store_type in ("itar", "directory") for zarr_format in (2, 3)
+]
+
+
+def _require_zarr_format(test: unittest.TestCase, zarr_format: int) -> None:
+    if zarr_format not in nodes.SUPPORTED_ZARR_FORMATS:
+        test.skipTest(f"zarr format v{zarr_format} requires zarr-python 3")
+
+
+@parameterized_class(("store_type", "zarr_format"), STORE_TYPES_AND_ZARR_FORMATS)
 class TestData4Reload(unittest.TestCase):
     """Test to verify functionality of V4 data writer + loader"""
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     def setUp(self):
+        _require_zarr_format(self, self.zarr_format)
         # Make printed errors more representable numerically
         np.set_printoptions(floatmode="unique", linewidth=200, suppress=True)
 
     @parameterized.expand(
         [
-            (False,),
-            (True,),
+            (False, None),
+            (True, None),
+            (True, 0),  # uncached nodes
+            (True, 1),  # nodes evicted on most accesses
         ]
     )
-    def test_reload(self, open_consolidated: bool):
+    def test_reload(self, open_consolidated: bool, node_cache_size: Optional[int]):
         """Test to make sure serialized data is faithfully reloaded"""
 
         tempdir = tempfile.TemporaryDirectory()
@@ -101,6 +115,7 @@ def test_reload(self, open_consolidated: bool):
                 ref_sequence_timestamp_interval_us := HalfClosedInterval(int(0 * 1e6), int(1 * 1e6) + 1)
             ),
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data=cast(Dict[str, JsonLike], ref_generic_sequence_meta_data := {"some": 1, "key": 1.2}),
         )
 
@@ -506,6 +521,7 @@ def normalize_points(vectors: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
             store_base_name=store_writer._store_base_name,
             sequence_reader=SequenceComponentGroupsReader(store_paths, open_consolidated=open_consolidated),
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
         )
 
         # Store cuboids with the new writer
@@ -573,7 +589,9 @@ def normalize_points(vectors: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
         store_paths += store_writer.finalize()
 
         ## Reload sequence and verify consistency
-        store_reader = SequenceComponentGroupsReader(store_paths, open_consolidated=open_consolidated)
+        store_reader = SequenceComponentGroupsReader(
+            store_paths, open_consolidated=open_consolidated, node_cache_size=node_cache_size
+        )
 
         # check sequence data
         self.assertEqual(store_reader.sequence_id, ref_sequence_id)
@@ -1096,6 +1114,7 @@ def test_component_generic_data_roundtrip(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -1160,6 +1179,7 @@ def test_component_generic_data_backwards_compat(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -1205,6 +1225,7 @@ def test_component_generic_data_meta_overwrite(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -1254,6 +1275,7 @@ def test_component_generic_data_meta_only(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -1294,13 +1316,7 @@ def test_component_generic_data_meta_only(self) -> None:
         tempdir.cleanup()
 
 
-@parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
-)
+@parameterized_class(("store_type", "zarr_format"), STORE_TYPES_AND_ZARR_FORMATS)
 class TestDataNewComponent(unittest.TestCase):
     """
     Test to demonstrate how to extend an existing dataset with a new custom component.
@@ -1312,8 +1328,10 @@ class TestDataNewComponent(unittest.TestCase):
     """
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     def setUp(self):
+        _require_zarr_format(self, self.zarr_format)
         np.set_printoptions(floatmode="unique", linewidth=200, suppress=True)
 
     def test_new_component_extension(self):
@@ -1342,6 +1360,7 @@ def test_new_component_extension(self):
             sequence_id=sequence_id,
             sequence_timestamp_interval_us=(timestamp_interval := HalfClosedInterval(int(0 * 1e6), int(10 * 1e6) + 1)),
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={"dataset": "test", "version": 1.0},
         )
 
@@ -1413,21 +1432,23 @@ def store_velocity(
                     return self
 
                 def finalize(self):
-                    """Write all velocity data to zarr storage"""
+                    """Write all velocity data to the component group"""
                     if self.velocities:
                         velocities_array = np.stack(self.velocities)
                         timestamps_array = np.array(self.timestamps, dtype=np.uint64)
 
-                        # Store as zarr arrays
-                        self._group.create_dataset(
+                        # Store as arrays
+                        self._group.create_array(
                             "velocities",
-                            data=velocities_array,
+                            velocities_array,
                             dtype=velocities_array.dtype,
+                            compression=BloscCompression(shuffle="shuffle"),
                         )
-                        self._group.create_dataset(
+                        self._group.create_array(
                             "timestamps_us",
-                            data=timestamps_array,
+                            timestamps_array,
                             dtype=np.uint64,
+                            compression=BloscCompression(shuffle="shuffle"),
                         )
 
             class Reader(ComponentReader):
@@ -1444,8 +1465,8 @@ def supports_component_version(version: str) -> bool:
 
                 def get_velocities(self) -> Tuple[np.ndarray, np.ndarray]:
                     """Returns (velocities, timestamps_us) arrays"""
-                    velocities = np.array(self._group["velocities"][:])
-                    timestamps_us = np.array(self._group["timestamps_us"][:])
+                    velocities = self._group.array("velocities").read()
+                    timestamps_us = self._group.array("timestamps_us").read()
                     return velocities, timestamps_us
 
         # ============================================================================
@@ -1471,6 +1492,7 @@ def get_velocities(self) -> Tuple[np.ndarray, np.ndarray]:
             output_dir_path=UPath(tempdir.name),  # Same directory as initial
             store_base_name=sequence_id + "-extension",
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
         )
 
         # Now add our custom velocity component to the extended dataset
@@ -1605,9 +1627,10 @@ def finalize(self):
                         accelerations_array = np.stack(self.accelerations)
                         timestamps_array = np.array(self.timestamps, dtype=np.uint64)
 
-                        self._group.create_dataset("velocities", data=velocities_array)
-                        self._group.create_dataset("accelerations", data=accelerations_array)  # NEW
-                        self._group.create_dataset("timestamps_us", data=timestamps_array)
+                        compression = BloscCompression(shuffle="shuffle")
+                        self._group.create_array("velocities", velocities_array, compression=compression)
+                        self._group.create_array("accelerations", accelerations_array, compression=compression)  # NEW
+                        self._group.create_array("timestamps_us", timestamps_array, compression=compression)
 
         # Create a backward-compatible reader that can read both v1 and v2
         class VelocityComponentBackwardCompatibleReader(ComponentReader):
@@ -1624,15 +1647,15 @@ def supports_component_version(version: str) -> bool:
 
             def get_velocities(self) -> Tuple[np.ndarray, np.ndarray]:
                 """Returns velocities (works for both v1 and v2)"""
-                velocities = np.array(self._group["velocities"][:])
-                timestamps_us = np.array(self._group["timestamps_us"][:])
+                velocities = self._group.array("velocities").read()
+                timestamps_us = self._group.array("timestamps_us").read()
                 return velocities, timestamps_us
 
             def get_accelerations(self) -> np.ndarray:
                 """Returns accelerations (only available in v2)"""
                 if self.component_version == "v1":
                     raise ValueError("Acceleration data not available in v1")
-                return np.array(self._group["accelerations"][:])
+                return self._group.array("accelerations").read()
 
         # Test that v1 reader cannot read v2 data
         v2_store_writer = SequenceComponentGroupsWriter(
@@ -1641,6 +1664,7 @@ def get_accelerations(self) -> np.ndarray:
             sequence_id="test_v2",
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={"version": "v2_test"},
         )
 
@@ -1704,19 +1728,15 @@ def get_accelerations(self) -> np.ndarray:
         # Version compatibility tests passed - all tests completed successfully
 
 
-@parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
-)
+@parameterized_class(("store_type", "zarr_format"), STORE_TYPES_AND_ZARR_FORMATS)
 class TestPointCloudsComponent(unittest.TestCase):
     """Round-trip tests for the PointCloudsComponent Writer/Reader."""
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     def setUp(self):
+        _require_zarr_format(self, self.zarr_format)
         np.set_printoptions(floatmode="unique", linewidth=200, suppress=True)
 
     def _make_writer_reader(
@@ -1735,6 +1755,7 @@ def _make_writer_reader(
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2059,19 +2080,15 @@ def test_generic_data_and_metadata(self):
         tmpdir.cleanup()
 
 
-@parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
-)
+@parameterized_class(("store_type", "zarr_format"), STORE_TYPES_AND_ZARR_FORMATS)
 class TestCameraLabelsComponent(unittest.TestCase):
     """Round-trip tests for the CameraLabelsComponent Writer/Reader."""
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     def setUp(self):
+        _require_zarr_format(self, self.zarr_format)
         np.set_printoptions(floatmode="unique", linewidth=200, suppress=True)
 
     def _make_writer(
@@ -2095,6 +2112,7 @@ def _make_writer(
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2420,6 +2438,7 @@ def test_multiple_label_types_per_camera(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type="directory",
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2582,6 +2601,7 @@ def test_reject_empty_camera_id(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2748,24 +2768,116 @@ def test_schema_json_roundtrip(self) -> None:
         self.assertIsNone(rt.quantization)
 
 
+class TestReaderOpening(unittest.TestCase):
+    """Component stores are opened concurrently, their hierarchies sequentially on the calling thread"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        writer = SequenceComponentGroupsWriter(
+            output_dir_path=UPath(self._tmp.name),
+            store_base_name="seq",
+            sequence_id="seq",
+            sequence_timestamp_interval_us=HalfClosedInterval(0, 10_000_001),
+            generic_meta_data={},
+        )
+        for name in ("a", "b", "c"):
+            writer.register_component_writer(PosesComponent.Writer, name, f"poses_{name}").store_static_pose(
+                "x", "y", np.eye(4, dtype=np.float32)
+            )
+        self.store_paths = writer.finalize()
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def test_hierarchies_are_opened_on_the_calling_thread(self):
+        threads = []
+        open_store = stores.open_store
+
+        def recording_open_store(*args: Any, **kwargs: Any) -> Any:
+            threads.append(threading.current_thread())
+            return open_store(*args, **kwargs)
+
+        with mock.patch.object(stores, "open_store", side_effect=recording_open_store):
+            reader = SequenceComponentGroupsReader(self.store_paths, max_threads=4)
+        self.assertEqual(len(threads), len(self.store_paths))
+        self.assertEqual(set(threads), {threading.current_thread()})
+        self.assertEqual(sorted(reader.open_component_readers(PosesComponent.Reader)), ["a", "b", "c"])
+
+
+class TestReaderNodeCaches(unittest.TestCase):
+    """Component readers cache the opened nodes of their component (sized by node_cache_size)"""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        writer = SequenceComponentGroupsWriter(
+            output_dir_path=UPath(self._tmp.name),
+            store_base_name="seq",
+            sequence_id="seq",
+            sequence_timestamp_interval_us=HalfClosedInterval(0, 10_000_001),
+            generic_meta_data={},
+        )
+        lidar_writer = writer.register_component_writer(LidarSensorComponent.Writer, "lidar", "lidar")
+        for i in range(4):
+            direction = np.tile(np.array([1.0, 0.0, 0.0], dtype=np.float32), (5, 1))
+            lidar_writer.store_frame(
+                direction=direction,
+                timestamp_us=np.full(5, 10 + i * 1_000_000, dtype=np.uint64),
+                model_element=None,
+                distance_m=np.full((1, 5), i, dtype=np.float32),
+                intensity=np.ones((1, 5), dtype=np.float32),
+                frame_timestamps_us=np.array([10 + i * 1_000_000, 20 + i * 1_000_000], dtype=np.uint64),
+                generic_data={},
+                generic_meta_data={},
+            )
+        self.store_paths = writer.finalize()
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    @parameterized.expand([(None,), (0,), (1,), (3,)])
+    def test_node_cache_sizes(self, node_cache_size: Optional[int]):
+        reader = SequenceComponentGroupsReader(self.store_paths, node_cache_size=node_cache_size)
+        lidar = reader.open_component_readers(LidarSensorComponent.Reader)["lidar"]
+        frames = [int(ts) for ts in lidar.frames_timestamps_us[:, 1]]
+        for _ in range(2):
+            for i, ts in enumerate(frames):
+                np.testing.assert_array_equal(lidar.get_frame_ray_bundle_return_data(ts, "distance_m", 0), [i] * 5)
+
+        cache = lidar._group.cache
+        if node_cache_size == 0:
+            self.assertIsNone(cache)
+            return
+        assert cache is not None
+        self.assertEqual(cache.max_entries, node_cache_size)
+        # every frame read opens the frame's returns group and its distance_m array
+        accessed = ["lidars/lidar/frames"] + [
+            f"lidars/lidar/frames/{ts}/ray_bundle_returns{suffix}" for ts in frames for suffix in ("", "/distance_m")
+        ]
+        if node_cache_size is None:
+            self.assertEqual(sorted(cache.keys()), sorted(accessed))
+        else:
+            # the most recently used nodes only
+            self.assertEqual(cache.keys(), accessed[-node_cache_size:])
+
+    def test_invalid_node_cache_size(self):
+        with self.assertRaises(ValueError):
+            SequenceComponentGroupsReader(self.store_paths, node_cache_size=-1)
+
+
 # ==============================================================================
-# Tests for zero-dimension array support (_normalize_chunks)
+# Tests for zero-dimension array support
 # ==============================================================================
 
 
-@parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
-)
+@parameterized_class(("store_type", "zarr_format"), STORE_TYPES_AND_ZARR_FORMATS)
 class TestZeroDimArraySupport(unittest.TestCase):
     """Tests that arrays with zero-length dimensions can be stored and read back correctly."""
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     def setUp(self):
+        _require_zarr_format(self, self.zarr_format)
         np.set_printoptions(floatmode="unique", linewidth=200, suppress=True)
 
     # ------------------------------------------------------------------
@@ -2782,6 +2894,7 @@ def test_point_cloud_zero_points(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2845,6 +2958,7 @@ def test_point_cloud_zero_points_multidim_generic_data(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2887,6 +3001,7 @@ def test_lidar_zero_rays(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -2953,6 +3068,7 @@ def test_radar_zero_rays(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -3009,6 +3125,7 @@ def test_camera_label_zero_shape_suffix_dim(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -3062,6 +3179,7 @@ def test_component_generic_data_zero_dim(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
@@ -3136,6 +3254,7 @@ def test_per_frame_generic_data_zero_dim(self) -> None:
             sequence_id=seq_id,
             sequence_timestamp_interval_us=timestamp_interval,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
diff --git a/ncore/impl/data/v4/formats_test.py b/ncore/impl/data/v4/formats_test.py
new file mode 100644
index 00000000..a109ab26
--- /dev/null
+++ b/ncore/impl/data/v4/formats_test.py
@@ -0,0 +1,394 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Cross-format / cross-runtime consistency of NCore V4 test data
+
+Compares the zarr format v2 and zarr format v3 serializations of the same (time-trimmed) test sequence, as well as
+re-serializations performed by the zarr-python version under test.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import tempfile
+import unittest
+
+from typing import Any, Dict, List, Literal, Optional, cast
+
+import numpy as np
+import parameterized
+
+from python.runfiles import Runfiles
+from upath import UPath
+
+from ncore.impl.data import nodes
+from ncore.impl.data.v4.components import (
+    CameraSensorComponent,
+    CuboidsComponent,
+    IntrinsicsComponent,
+    LidarSensorComponent,
+    MasksComponent,
+    PosesComponent,
+    RadarSensorComponent,
+    SequenceComponentGroupsReader,
+    SequenceComponentGroupsWriter,
+)
+
+
+_RUNFILES = Runfiles.Create()
+assert _RUNFILES is not None
+_RUNFILES_: Runfiles = _RUNFILES
+
+_SEQUENCE = "c9b05cf4-afb9-11ec-b3c2-00044bf65fcb@1648597318700123-1648599151600035"
+
+
+def _dataset(name: str) -> UPath:
+    path = UPath(_RUNFILES_.Rlocation(f"test-data-v4-trimmed/{name}/{_SEQUENCE}.json") or "")
+    assert path.exists(), f"test data {name} not found (missing data dependency?)"
+    return path
+
+
+_TRIMMED_V2 = _dataset("zarr-v2")
+_TRIMMED_V3 = _dataset("zarr-v3")
+
+
+#: Frame subsampling of the reserialization test (every N-th frame, plus the last frame of each sensor)
+_FRAME_STRIDE = 5
+
+
+def _subsampled(frames_timestamps_us: np.ndarray) -> np.ndarray:
+    """Subsampled frames (every _FRAME_STRIDE-th frame, plus the last frame)"""
+    indices = sorted(set(range(0, len(frames_timestamps_us), _FRAME_STRIDE)) | {len(frames_timestamps_us) - 1})
+    return frames_timestamps_us[[i for i in indices if i >= 0]]
+
+
+def _summary(reader: SequenceComponentGroupsReader, subsampled: bool = False, frames: bool = True) -> Dict[str, Any]:
+    """Collects a comparable summary of all data of a sequence (optionally of the subsampled frames only, or without
+    masks and sensor frames)"""
+
+    def selected(timestamps_us: np.ndarray) -> np.ndarray:
+        return _subsampled(timestamps_us) if subsampled else timestamps_us
+
+    out: Dict[str, Any] = {
+        "sequence_id": reader.sequence_id,
+        "interval": (reader.sequence_timestamp_interval_us.start, reader.sequence_timestamp_interval_us.stop),
+        "generic_meta_data": reader.generic_meta_data,
+    }
+
+    for name, poses in sorted(reader.open_component_readers(PosesComponent.Reader).items()):
+        out[f"poses/{name}/static"] = {k: v.tolist() for k, v in poses.get_static_poses()}
+        out[f"poses/{name}/dynamic"] = {k: (p.tolist(), t.tolist()) for k, (p, t) in poses.get_dynamic_poses()}
+
+    for name, intrinsics in sorted(reader.open_component_readers(IntrinsicsComponent.Reader).items()):
+        out[f"intrinsics/{name}"] = (intrinsics._group.group("cameras").attrs, intrinsics._group.group("lidars").attrs)
+
+    for name, cuboids in sorted(reader.open_component_readers(CuboidsComponent.Reader).items()):
+        out[f"cuboids/{name}"] = [obs.to_dict() for obs in cuboids.get_observations()]
+
+    if not frames:
+        return out
+
+    for name, masks in sorted(reader.open_component_readers(MasksComponent.Reader).items()):
+        out[f"masks/{name}"] = {
+            camera_id: {
+                mask_name: np.asarray(image).tobytes() for mask_name, image in masks.get_camera_mask_images(camera_id)
+            }
+            for camera_id in masks._group.group("cameras").members()
+        }
+
+    for name, cameras in sorted(reader.open_component_readers(CameraSensorComponent.Reader).items()):
+        frames: List[object] = []
+        for ts in selected(cameras.frames_timestamps_us)[:, 1]:
+            ts = int(ts)
+            data = cameras.get_frame_data(ts)
+            frames.append(
+                (
+                    ts,
+                    data.get_encoded_image_format(),
+                    data.get_encoded_image_data(),
+                    {
+                        n: cameras.get_frame_generic_data(ts, n).tobytes()
+                        for n in sorted(cameras.get_frame_generic_data_names(ts))
+                    },
+                    cameras.get_frame_generic_meta_data(ts),
+                )
+            )
+        out[f"cameras/{name}"] = frames
+
+    for component in (LidarSensorComponent, RadarSensorComponent):
+        for name, sensor in sorted(reader.open_component_readers(component.Reader).items()):
+            sensor_frames = []
+            for ts in selected(sensor.frames_timestamps_us)[:, 1]:
+                ts = int(ts)
+                sensor_frames.append(
+                    (
+                        ts,
+                        {
+                            n: sensor.get_frame_ray_bundle_data(ts, n).tobytes()
+                            for n in sorted(sensor.get_frame_ray_bundle_data_names(ts))
+                        },
+                        {
+                            n: sensor.get_frame_ray_bundle_return_data(ts, n, None).tobytes()
+                            for n in sorted(sensor.get_frame_ray_bundle_return_data_names(ts))
+                        },
+                        sensor.get_frame_ray_bundle_return_valid_mask(ts).tobytes(),
+                        {
+                            n: sensor.get_frame_generic_data(ts, n).tobytes()
+                            for n in sorted(sensor.get_frame_generic_data_names(ts))
+                        },
+                        sensor.get_frame_generic_meta_data(ts),
+                    )
+                )
+            out[f"{component.COMPONENT_NAME}/{name}"] = sensor_frames
+
+    return out
+
+
+async def _summary_async(reader: SequenceComponentGroupsReader) -> Dict[str, Any]:
+    """:func:`_summary` via the asynchronous reader APIs, all frames of all sensors read concurrently"""
+    out = _summary(reader, subsampled=False, frames=False)
+
+    async def masks(masks: MasksComponent.Reader) -> Dict[str, Any]:
+        out: Dict[str, Any] = {}
+        for camera_id in masks._group.group("cameras").members():
+            out[camera_id] = {}
+            for mask_name in masks.get_camera_mask_names(camera_id):
+                image = await masks.get_camera_mask_image_async(camera_id, mask_name)
+                out[camera_id][mask_name] = np.asarray(image).tobytes()
+        return out
+
+    async def camera_frame(cameras: CameraSensorComponent.Reader, ts: int) -> object:
+        data, generic = await asyncio.gather(
+            cameras.get_frame_data_async(ts),
+            asyncio.gather(
+                *(cameras.get_frame_generic_data_async(ts, n) for n in sorted(cameras.get_frame_generic_data_names(ts)))
+            ),
+        )
+        names = sorted(cameras.get_frame_generic_data_names(ts))
+        return (
+            ts,
+            data.get_encoded_image_format(),
+            data.get_encoded_image_data(),
+            {n: value.tobytes() for n, value in zip(names, generic)},
+            cameras.get_frame_generic_meta_data(ts),
+        )
+
+    async def sensor_frame(sensor: Any, ts: int) -> object:
+        ray_names = sorted(sensor.get_frame_ray_bundle_data_names(ts))
+        return_names = sorted(sensor.get_frame_ray_bundle_return_data_names(ts))
+        generic_names = sorted(sensor.get_frame_generic_data_names(ts))
+        rays, returns, mask, generic = await asyncio.gather(
+            asyncio.gather(*(sensor.get_frame_ray_bundle_data_async(ts, n) for n in ray_names)),
+            asyncio.gather(*(sensor.get_frame_ray_bundle_return_data_async(ts, n, None) for n in return_names)),
+            sensor.get_frame_ray_bundle_return_valid_mask_async(ts),
+            asyncio.gather(*(sensor.get_frame_generic_data_async(ts, n) for n in generic_names)),
+        )
+        return (
+            ts,
+            {n: v.tobytes() for n, v in zip(ray_names, rays)},
+            {n: v.tobytes() for n, v in zip(return_names, returns)},
+            mask.tobytes(),
+            {n: v.tobytes() for n, v in zip(generic_names, generic)},
+            sensor.get_frame_generic_meta_data(ts),
+        )
+
+    for name, masks_reader in sorted(reader.open_component_readers(MasksComponent.Reader).items()):
+        out[f"masks/{name}"] = await masks(masks_reader)
+    for name, cameras in sorted(reader.open_component_readers(CameraSensorComponent.Reader).items()):
+        out[f"cameras/{name}"] = list(
+            await asyncio.gather(*(camera_frame(cameras, int(ts)) for ts in cameras.frames_timestamps_us[:, 1]))
+        )
+    for component in (LidarSensorComponent, RadarSensorComponent):
+        for name, sensor in sorted(reader.open_component_readers(component.Reader).items()):
+            out[f"{component.COMPONENT_NAME}/{name}"] = list(
+                await asyncio.gather(*(sensor_frame(sensor, int(ts)) for ts in sensor.frames_timestamps_us[:, 1]))
+            )
+    return out
+
+
+class TestZarrFormats(unittest.TestCase):
+    """zarr format v2 and v3 serializations of a sequence need to be identical in content"""
+
+    reference: Dict[str, Any]  # summary of the zarr format v2 test data
+    reference_subsampled: Dict[str, Any]  # summary of the subsampled frames of the zarr format v2 test data
+
+    @classmethod
+    def setUpClass(cls) -> None:
+        cls.reference = _summary(SequenceComponentGroupsReader([_TRIMMED_V2]))
+        cls.reference_subsampled = _summary(SequenceComponentGroupsReader([_TRIMMED_V2]), subsampled=True)
+
+    def test_v2_v3_consistency(self):
+        if 3 not in nodes.SUPPORTED_ZARR_FORMATS:
+            # zarr format v3 data can't be read with zarr-python 2
+            with self.assertRaises(Exception):
+                SequenceComponentGroupsReader([_TRIMMED_V3])
+            return
+
+        _assert_summaries_equal(self, self.reference, _summary(SequenceComponentGroupsReader([_TRIMMED_V3])))
+
+    @parameterized.parameterized.expand([(None,), (0,), (1,), (16,)])
+    def test_node_cache_sizes(self, node_cache_size: Optional[int]):
+        """All node cache configurations of the readers return the same data, and caches respect their bounds"""
+        reader = SequenceComponentGroupsReader([_TRIMMED_V2], node_cache_size=node_cache_size)
+        _assert_summaries_equal(self, self.reference, _summary(reader))
+
+        lidar = next(iter(reader.open_component_readers(LidarSensorComponent.Reader).values()))
+        frames = lidar.frames_timestamps_us[:, 1]
+        for _ in range(2):
+            for ts in frames:
+                lidar.get_frame_ray_bundle_data(int(ts), "direction")
+        if node_cache_size == 0:
+            self.assertIsNone(lidar._group.cache)
+        else:
+            assert (cache := lidar._group.cache) is not None
+            self.assertEqual(cache.max_entries, node_cache_size)
+            self.assertLessEqual(len(cache), node_cache_size or len(cache))
+            frame_entries = [key for key in cache.keys() if key.endswith("/ray_bundle/direction")]
+            self.assertEqual(len(frame_entries), min(len(frames), node_cache_size or len(frames)))
+
+    @parameterized.parameterized.expand([("v2",), ("v3",)])
+    def test_async_reads(self, data: str):
+        """Asynchronous reader APIs (all frames read concurrently) return the same data as the synchronous APIs"""
+        if data == "v3" and 3 not in nodes.SUPPORTED_ZARR_FORMATS:
+            self.skipTest("zarr format v3 data requires zarr-python 3")
+        reader = SequenceComponentGroupsReader([_TRIMMED_V2 if data == "v2" else _TRIMMED_V3])
+        _assert_summaries_equal(self, self.reference, asyncio.run(_summary_async(reader)))
+
+    def test_invalid_node_cache_size(self):
+        with self.assertRaises(ValueError):
+            SequenceComponentGroupsReader([_TRIMMED_V2], node_cache_size=-1)
+
+    @parameterized.parameterized.expand([(f,) for f in nodes.SUPPORTED_ZARR_FORMATS])
+    def test_reserialization(self, zarr_format: int):
+        """Re-serializing (subsampled frames of) all components of the test data with the installed zarr-python version
+        via the public component APIs preserves all content"""
+        reader = SequenceComponentGroupsReader([_TRIMMED_V2])
+
+        with tempfile.TemporaryDirectory() as output_dir:
+            writer = SequenceComponentGroupsWriter.from_reader(
+                UPath(output_dir), "copy", reader, zarr_format=cast(Literal[2, 3], zarr_format)
+            )
+            _copy_components(reader, writer)
+            copy = _summary(SequenceComponentGroupsReader(writer.finalize()))  # (contains the subsampled frames only)
+
+        self.assertEqual(self.reference_subsampled.keys(), copy.keys())
+        for component in ("poses/", "intrinsics/", "masks/", "cameras/", "lidars/", "radars/", "cuboids/"):
+            self.assertTrue(any(key.startswith(component) for key in copy), f"no {component} data in the test data")
+        _assert_summaries_equal(self, self.reference_subsampled, copy)
+
+
+def _assert_summaries_equal(test: unittest.TestCase, expected: Dict[str, Any], actual: Dict[str, Any]) -> None:
+    """Compares sequence summaries per component / frame / field (without expensive diffs of large payloads)"""
+    test.assertEqual(sorted(expected), sorted(actual))
+    for key, expected_value in expected.items():
+        actual_value = actual[key]
+        if not isinstance(expected_value, list):
+            test.assertTrue(expected_value == actual_value, key)
+            continue
+        test.assertEqual(len(expected_value), len(actual_value), key)
+        for index, (expected_frame, actual_frame) in enumerate(zip(expected_value, actual_value)):
+            if not isinstance(expected_frame, tuple):
+                test.assertTrue(expected_frame == actual_frame, f"{key}[{index}]")
+                continue
+            for field, (expected_field, actual_field) in enumerate(zip(expected_frame, actual_frame)):
+                if isinstance(expected_field, dict):
+                    expected_items = cast(Dict[str, object], expected_field)
+                    actual_items = cast(Dict[str, object], actual_field)
+                    test.assertEqual(sorted(expected_items), sorted(actual_items), f"{key}[{index}][{field}]")
+                    for name, expected_item in expected_items.items():
+                        test.assertTrue(expected_item == actual_items[name], f"{key}[{index}][{field}][{name}] differs")
+                else:
+                    test.assertTrue(expected_field == actual_field, f"{key}[{index}][{field}] differs")
+
+
+def _copy_components(reader: SequenceComponentGroupsReader, writer: SequenceComponentGroupsWriter) -> None:
+    """Copies (subsampled frames of) all components of a sequence via the public component APIs"""
+
+    for name, poses in reader.open_component_readers(PosesComponent.Reader).items():
+        poses_writer = writer.register_component_writer(PosesComponent.Writer, name, name, poses.generic_meta_data)
+        for (source, target), pose in poses.get_static_poses():
+            poses_writer.store_static_pose(source, target, pose)
+        for (source, target), (trajectory, timestamps_us) in poses.get_dynamic_poses():
+            poses_writer.store_dynamic_pose(source, target, trajectory, timestamps_us)
+
+    for name, intrinsics in reader.open_component_readers(IntrinsicsComponent.Reader).items():
+        intrinsics_writer = writer.register_component_writer(
+            IntrinsicsComponent.Writer, name, name, intrinsics.generic_meta_data
+        )
+        for camera_id in intrinsics._group.group("cameras").members():
+            intrinsics_writer.store_camera_intrinsics(camera_id, intrinsics.get_camera_model_parameters(camera_id))
+        for lidar_id in intrinsics._group.group("lidars").members():
+            if (lidar_model := intrinsics.get_lidar_model_parameters(lidar_id)) is not None:
+                intrinsics_writer.store_lidar_intrinsics(lidar_id, lidar_model)
+
+    for name, masks in reader.open_component_readers(MasksComponent.Reader).items():
+        masks_writer = writer.register_component_writer(MasksComponent.Writer, name, name, masks.generic_meta_data)
+        for camera_id in masks._group.group("cameras").members():
+            masks_writer.store_camera_masks(camera_id, dict(masks.get_camera_mask_images(camera_id)))
+
+    for name, cameras in reader.open_component_readers(CameraSensorComponent.Reader).items():
+        camera_writer = writer.register_component_writer(
+            CameraSensorComponent.Writer, name, name, cameras.generic_meta_data
+        )
+        for frame_timestamps_us in _subsampled(cameras.frames_timestamps_us):
+            ts = int(frame_timestamps_us[1])
+            data = cameras.get_frame_data(ts)
+            camera_writer.store_frame(
+                data.get_encoded_image_data(),
+                data.get_encoded_image_format(),
+                frame_timestamps_us,
+                {n: cameras.get_frame_generic_data(ts, n) for n in cameras.get_frame_generic_data_names(ts)},
+                cameras.get_frame_generic_meta_data(ts),
+            )
+
+    for name, lidar in reader.open_component_readers(LidarSensorComponent.Reader).items():
+        lidar_writer = writer.register_component_writer(
+            LidarSensorComponent.Writer, name, name, lidar.generic_meta_data
+        )
+        for frame_timestamps_us in _subsampled(lidar.frames_timestamps_us):
+            ts = int(frame_timestamps_us[1])
+            ray_names = lidar.get_frame_ray_bundle_data_names(ts)
+            lidar_writer.store_frame(
+                direction=lidar.get_frame_ray_bundle_data(ts, "direction"),
+                timestamp_us=lidar.get_frame_ray_bundle_data(ts, "timestamp_us"),
+                model_element=(
+                    lidar.get_frame_ray_bundle_data(ts, "model_element") if "model_element" in ray_names else None
+                ),
+                distance_m=lidar.get_frame_ray_bundle_return_data(ts, "distance_m", None),
+                intensity=lidar.get_frame_ray_bundle_return_data(ts, "intensity", None),
+                frame_timestamps_us=frame_timestamps_us,
+                generic_data={n: lidar.get_frame_generic_data(ts, n) for n in lidar.get_frame_generic_data_names(ts)},
+                generic_meta_data=lidar.get_frame_generic_meta_data(ts),
+            )
+
+    for name, radar in reader.open_component_readers(RadarSensorComponent.Reader).items():
+        radar_writer = writer.register_component_writer(
+            RadarSensorComponent.Writer, name, name, radar.generic_meta_data
+        )
+        for frame_timestamps_us in _subsampled(radar.frames_timestamps_us):
+            ts = int(frame_timestamps_us[1])
+            radar_writer.store_frame(
+                direction=radar.get_frame_ray_bundle_data(ts, "direction"),
+                timestamp_us=radar.get_frame_ray_bundle_data(ts, "timestamp_us"),
+                distance_m=radar.get_frame_ray_bundle_return_data(ts, "distance_m", None),
+                frame_timestamps_us=frame_timestamps_us,
+                generic_data={n: radar.get_frame_generic_data(ts, n) for n in radar.get_frame_generic_data_names(ts)},
+                generic_meta_data=radar.get_frame_generic_meta_data(ts),
+            )
+
+    for name, cuboids in reader.open_component_readers(CuboidsComponent.Reader).items():
+        writer.register_component_writer(
+            CuboidsComponent.Writer, name, name, cuboids.generic_meta_data
+        ).store_observations(list(cuboids.get_observations()))
diff --git a/ncore/impl/data_converter/base.py b/ncore/impl/data_converter/base.py
index 7a692bf6..347a813b 100644
--- a/ncore/impl/data_converter/base.py
+++ b/ncore/impl/data_converter/base.py
@@ -20,7 +20,7 @@
 
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
-from typing import Optional, Tuple
+from typing import Literal, Optional, Tuple
 
 from upath import UPath
 
@@ -45,6 +45,9 @@ class BaseDataConverterConfig:
     debug: bool
     debug_port: int
 
+    ## Output
+    zarr_format: Literal[2, 3]  # On-disk zarr format of written component stores (3 requires zarr-python 3)
+
 
 class BaseDataConverter(ABC):
     """
@@ -62,6 +65,9 @@ def __init__(self, config: BaseDataConverterConfig) -> None:
 
         self.output_dir = UPath(config.output_dir)
 
+        # On-disk zarr format of written component stores
+        self.zarr_format: Literal[2, 3] = config.zarr_format
+
         # External sensor selection overwrites
         # Store `None`` for `_active_<sensor>_ids` in case all sensors should be used, as the
         # actual full list of sensor ids will be passed via `get_active_<sensor>_ids()`
diff --git a/tools/data_converter/argoverse2/BUILD.bazel b/tools/data_converter/argoverse2/BUILD.bazel
index a645eb51..cf67b628 100644
--- a/tools/data_converter/argoverse2/BUILD.bazel
+++ b/tools/data_converter/argoverse2/BUILD.bazel
@@ -67,6 +67,7 @@ alias(
 # Integration test for the Argoverse 2 converter (requires AV2_DIR env var)
 pytest_test(
     name = "pytest_converter",
+    size = "large",  # converts the test sequence for all store types / zarr formats
     srcs = ["converter_test.py"],
     python_versions = ["3.11"],
     tags = ["manual"],  # Only run when explicitly requested (needs external data)
diff --git a/tools/data_converter/argoverse2/converter.py b/tools/data_converter/argoverse2/converter.py
index 7c999e1a..ceba5c5a 100644
--- a/tools/data_converter/argoverse2/converter.py
+++ b/tools/data_converter/argoverse2/converter.py
@@ -245,6 +245,7 @@ def convert_sequence(self, sequence_id: str) -> None:
             sequence_id=log_id,
             sequence_timestamp_interval_us=sequence_timestamp_interval_us,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={
                 "source_dataset": "argoverse2",
                 "argoverse2_split": self._split,
diff --git a/tools/data_converter/argoverse2/converter_test.py b/tools/data_converter/argoverse2/converter_test.py
index c70bc25e..e76c7fca 100644
--- a/tools/data_converter/argoverse2/converter_test.py
+++ b/tools/data_converter/argoverse2/converter_test.py
@@ -33,6 +33,7 @@
 from parameterized import parameterized_class
 from upath import UPath
 
+from ncore.impl.data import nodes
 from ncore.impl.data.types import (
     IdealPinholeCameraModelParameters,
     RowOffsetStructuredSpinningLidarModelParameters,
@@ -52,11 +53,8 @@
 
 
 @parameterized_class(
-    ("store_type",),
-    [
-        ("itar",),
-        ("directory",),
-    ],
+    ("store_type", "zarr_format"),
+    [(store_type, zarr_format) for store_type in ("itar", "directory") for zarr_format in nodes.SUPPORTED_ZARR_FORMATS],
 )
 class TestArgoverse2Converter(unittest.TestCase):
     """Integration tests for the Argoverse 2 data converter.
@@ -66,6 +64,7 @@ class TestArgoverse2Converter(unittest.TestCase):
     """
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     @classmethod
     def setUpClass(cls):
@@ -97,6 +96,7 @@ def setUpClass(cls):
             split=cls.split,
             log_id=cls.log_id,
             store_type=cls.store_type,
+            zarr_format=cls.zarr_format,
             component_group_profile="separate-sensors",
             store_sequence_meta=True,
         )
@@ -284,7 +284,7 @@ def test_lidar_model_reconstructs_directions(self):
             ts = int(reader.frames_timestamps_us[0, 1])
             direction = reader.get_frame_ray_bundle_data(ts, "direction")
             elem = reader.get_frame_ray_bundle_data(ts, "model_element")
-            distance = np.asarray(reader._get_ray_bundle_returns_group(ts)["distance_m"])[0]
+            distance = reader._get_ray_bundle_returns_group(ts).array("distance_m").read()[0]
 
             far = np.isfinite(distance) & (distance > 20.0) & (np.linalg.norm(direction, axis=1) > 0)
             self.assertGreater(int(far.sum()), 100, f"{unit}: too few far returns to validate")
@@ -352,7 +352,7 @@ def test_cuboids_align_with_lidar(self):
         rig_poses, pose_ts = poses_reader.get_dynamic_pose("rig", "world")
 
         direction = lidar_reader.get_frame_ray_bundle_data(ts, "direction")
-        distance = np.asarray(lidar_reader._get_ray_bundle_returns_group(ts)["distance_m"])[0]
+        distance = lidar_reader._get_ray_bundle_returns_group(ts).array("distance_m").read()[0]
         point_ts = lidar_reader.get_frame_ray_bundle_data(ts, "timestamp_us").astype(np.int64)
         valid = np.isfinite(distance) & (distance > 0)
         pts_sensor = direction[valid] * distance[valid, None]
diff --git a/tools/data_converter/cli.py b/tools/data_converter/cli.py
index 2e394587..89eb38e3 100644
--- a/tools/data_converter/cli.py
+++ b/tools/data_converter/cli.py
@@ -44,6 +44,13 @@
 @click.option(
     "--debug-port", type=int, default=5678, help="The port on which debugpy will wait for a client to connect"
 )
+@click.option(
+    "--zarr-format",
+    type=click.Choice(["2", "3"]),
+    default="2",
+    callback=lambda _ctx, _param, value: int(value),
+    help="On-disk zarr format of written component stores (format 3 requires zarr-python 3 for writing and reading)",
+)
 @click.option("--no-cameras", is_flag=True, default=False, help="Disable exporting any camera sensor")
 @click.option(
     "--camera-id",
diff --git a/tools/data_converter/colmap/converter.py b/tools/data_converter/colmap/converter.py
index 2cb28c7c..1e3e4d26 100644
--- a/tools/data_converter/colmap/converter.py
+++ b/tools/data_converter/colmap/converter.py
@@ -265,6 +265,7 @@ def convert_sequence(self, sequence_id: str) -> None:
             sequence_id=self.sequence_name,
             sequence_timestamp_interval_us=sequence_timestamp_interval_us,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data=self.generic_meta_data,
         )
 
@@ -768,6 +769,7 @@ def scannetpp_v4(ctx, **kwargs):
             lidar_ids=base.lidar_ids,
             no_radars=base.no_radars,
             radar_ids=base.radar_ids,
+            zarr_format=base.zarr_format,
             colmap_dir="dslr/colmap",
             images_dir=images_dir,
             masks_dir=masks_dir,
diff --git a/tools/data_converter/kitti/BUILD.bazel b/tools/data_converter/kitti/BUILD.bazel
index 04946c07..e9e8a5e3 100644
--- a/tools/data_converter/kitti/BUILD.bazel
+++ b/tools/data_converter/kitti/BUILD.bazel
@@ -59,6 +59,7 @@ alias(
 # Integration test for the KITTI converter (requires KITTI_RAW_DIR env var)
 pytest_test(
     name = "pytest_converter",
+    size = "large",  # converts the test sequence for all store types / zarr formats
     srcs = ["converter_test.py"],
     python_versions = ["3.11"],
     tags = ["manual"],  # Only run when explicitly requested (needs external data)
diff --git a/tools/data_converter/kitti/converter.py b/tools/data_converter/kitti/converter.py
index 39b464a3..919039af 100644
--- a/tools/data_converter/kitti/converter.py
+++ b/tools/data_converter/kitti/converter.py
@@ -211,6 +211,7 @@ def convert_sequence(self, sequence_id: str) -> None:
             sequence_id=sequence_name,
             sequence_timestamp_interval_us=sequence_timestamp_interval_us,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},
         )
 
diff --git a/tools/data_converter/kitti/converter_test.py b/tools/data_converter/kitti/converter_test.py
index 17121df6..5c59ae38 100644
--- a/tools/data_converter/kitti/converter_test.py
+++ b/tools/data_converter/kitti/converter_test.py
@@ -26,6 +26,7 @@
 from parameterized import parameterized_class
 from upath import UPath
 
+from ncore.impl.data import nodes
 from ncore.impl.data.types import IdealPinholeCameraModelParameters
 from ncore.impl.data.v4.components import (
     CameraSensorComponent,
@@ -39,11 +40,8 @@
 
 
 @parameterized_class(
-    ("store_type"),
-    [
-        ("itar",),
-        ("directory",),
-    ],
+    ("store_type", "zarr_format"),
+    [(store_type, zarr_format) for store_type in ("itar", "directory") for zarr_format in nodes.SUPPORTED_ZARR_FORMATS],
 )
 class TestKittiConverter(unittest.TestCase):
     """Integration tests for KITTI raw data converter.
@@ -54,6 +52,7 @@ class TestKittiConverter(unittest.TestCase):
     """
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     @classmethod
     def setUpClass(cls):
@@ -78,6 +77,7 @@ def setUpClass(cls):
             debug=False,
             debug_port=5678,
             store_type=cls.store_type,
+            zarr_format=cls.zarr_format,
             component_group_profile="separate-sensors",
             store_sequence_meta=True,
         )
diff --git a/tools/data_converter/nuscenes/BUILD.bazel b/tools/data_converter/nuscenes/BUILD.bazel
index e16cb101..8eb13151 100644
--- a/tools/data_converter/nuscenes/BUILD.bazel
+++ b/tools/data_converter/nuscenes/BUILD.bazel
@@ -65,6 +65,7 @@ alias(
 # Integration test for the nuScenes converter (requires NUSCENES_DIR env var)
 pytest_test(
     name = "pytest_converter",
+    size = "large",  # converts the test sequence for all store types / zarr formats
     srcs = ["converter_test.py"],
     python_versions = ["3.11"],
     tags = ["manual"],  # Only run when explicitly requested (needs external data)
diff --git a/tools/data_converter/nuscenes/converter.py b/tools/data_converter/nuscenes/converter.py
index e8b31915..0332a22b 100644
--- a/tools/data_converter/nuscenes/converter.py
+++ b/tools/data_converter/nuscenes/converter.py
@@ -280,6 +280,7 @@ def convert_sequence(self, sequence_id: str) -> None:
             sequence_id=sequence_output_name,
             sequence_timestamp_interval_us=sequence_timestamp_interval_us,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={
                 "source_dataset": "nuscenes",
                 "nuscenes_version": self._version,
diff --git a/tools/data_converter/nuscenes/converter_test.py b/tools/data_converter/nuscenes/converter_test.py
index 6e5edb83..649205e9 100644
--- a/tools/data_converter/nuscenes/converter_test.py
+++ b/tools/data_converter/nuscenes/converter_test.py
@@ -33,6 +33,7 @@
 from parameterized import parameterized_class
 from upath import UPath
 
+from ncore.impl.data import nodes
 from ncore.impl.data.types import IdealPinholeCameraModelParameters, RowOffsetStructuredSpinningLidarModelParameters
 from ncore.impl.data.v4.components import (
     CameraSensorComponent,
@@ -49,11 +50,8 @@
 
 
 @parameterized_class(
-    ("store_type",),
-    [
-        ("itar",),
-        ("directory",),
-    ],
+    ("store_type", "zarr_format"),
+    [(store_type, zarr_format) for store_type in ("itar", "directory") for zarr_format in nodes.SUPPORTED_ZARR_FORMATS],
 )
 class TestNuScenesConverter(unittest.TestCase):
     """Integration tests for nuScenes data converter.
@@ -63,6 +61,7 @@ class TestNuScenesConverter(unittest.TestCase):
     """
 
     store_type: Literal["itar", "directory"]
+    zarr_format: Literal[2, 3]
 
     @classmethod
     def setUpClass(cls):
@@ -97,6 +96,7 @@ def setUpClass(cls):
             scene_token=cls.scene_token,
             scene_name=None,
             store_type=cls.store_type,
+            zarr_format=cls.zarr_format,
             component_group_profile="separate-sensors",
             store_sequence_meta=True,
         )
@@ -294,7 +294,7 @@ def test_lidar_model_reproduces_point_cloud(self):
 
             # Native: direction * distance
             direction = lidar_reader.get_frame_ray_bundle_data(ts, "direction")
-            distance_2d = np.array(lidar_reader._get_ray_bundle_returns_group(ts)["distance_m"])  # [R, N]
+            distance_2d = lidar_reader._get_ray_bundle_returns_group(ts).array("distance_m").read()  # [R, N]
             distance = distance_2d[0]  # first return
 
             # Model element
diff --git a/tools/data_converter/pai/converter.py b/tools/data_converter/pai/converter.py
index f6bd419c..ba79f23e 100644
--- a/tools/data_converter/pai/converter.py
+++ b/tools/data_converter/pai/converter.py
@@ -175,6 +175,7 @@ def _init_pai_fields(self, config) -> None:
         self.seek_sec: float | None = config.seek_sec
         self.duration_sec: float | None = config.duration_sec
         self.store_type: Literal["itar", "directory"] = config.store_type
+        self.zarr_format: Literal[2, 3] = config.zarr_format
         self.component_group_profile: Literal["default", "separate-sensors", "separate-all"] = (
             config.component_group_profile
         )
@@ -286,6 +287,7 @@ def _convert_clip(self, clip_id: str) -> None:
             sequence_id=sequence_id,
             sequence_timestamp_interval_us=self.sequence_timestamp_interval_us,
             store_type=self.store_type,  # can also be "directory"
+            zarr_format=self.zarr_format,
             generic_meta_data={**source_generic_meta_data, **generic_meta_data},
         )
 
diff --git a/tools/data_converter/waymo/converter.py b/tools/data_converter/waymo/converter.py
index 50196768..2eb47649 100644
--- a/tools/data_converter/waymo/converter.py
+++ b/tools/data_converter/waymo/converter.py
@@ -243,6 +243,7 @@ def convert_sequence(self, sequence_id: str) -> None:
             sequence_id=sequence_name,
             sequence_timestamp_interval_us=sequence_timestamp_interval_us,
             store_type=self.store_type,
+            zarr_format=self.zarr_format,
             generic_meta_data={},  # no generic sequence meta data
         )