diff --git a/cuda_core/cuda/core/_device_resources.pyx b/cuda_core/cuda/core/_device_resources.pyx
index 0c01956d2ea..26c8863e063 100644
--- a/cuda_core/cuda/core/_device_resources.pyx
+++ b/cuda_core/cuda/core/_device_resources.pyx
@@ -21,6 +21,7 @@ from cuda.core._resource_handles cimport ContextHandle, GreenCtxHandle, as_cu, g
 from cuda.core._utils.cuda_utils cimport check_or_create_options, HANDLE_RETURN
 from cuda.core._utils.cuda_utils import is_sequence
 from cuda.core._utils.version cimport cy_binding_version, cy_driver_version
+from cuda.core._utils.validators import check_str_enum
 
 
 __all__ = [
@@ -147,11 +148,8 @@ cdef class WorkqueueResourceOptions:
     concurrency_limit: int | None = None
 
     def __post_init__(self):
-        if self.sharing_scope not in (None, "device_ctx", "green_ctx_balanced"):
-            raise ValueError(
-                f"Unknown sharing_scope: {self.sharing_scope!r}. "
-                "Expected 'device_ctx' or 'green_ctx_balanced'."
-            )
+        from cuda.core.typing import WorkqueueSharingScopeType
+        check_str_enum(self.sharing_scope, WorkqueueSharingScopeType, allow_none=True)
         if self.concurrency_limit is not None and self.concurrency_limit < 1:
             raise ValueError(
                 f"concurrency_limit must be >= 1, got {self.concurrency_limit}"
diff --git a/cuda_core/cuda/core/_memory/_managed_memory_resource.pyx b/cuda_core/cuda/core/_memory/_managed_memory_resource.pyx
index 41b7a992bed..d5a637a7b50 100644
--- a/cuda_core/cuda/core/_memory/_managed_memory_resource.pyx
+++ b/cuda_core/cuda/core/_memory/_managed_memory_resource.pyx
@@ -21,6 +21,9 @@ import warnings
 from cuda.core._memory._managed_buffer import ManagedBuffer
 from cuda.core.typing import ManagedMemoryLocationType
 
+IF CUDA_CORE_BUILD_MAJOR >= 13:
+    from cuda.core._utils.validators import check_str_enum
+
 if TYPE_CHECKING:
     from cuda.core.graph import GraphBuilder
 
@@ -163,9 +166,6 @@ cdef class ManagedMemoryResource(_MemPool):
 
 
 IF CUDA_CORE_BUILD_MAJOR >= 13:
-    cdef tuple _VALID_LOCATION_TYPES = ("device", "host", "host_numa")
-
-
     cdef _resolve_preferred_location(ManagedMemoryResourceOptions opts):
         """Resolve preferred location options into driver and stored values.
 
@@ -175,11 +175,7 @@ IF CUDA_CORE_BUILD_MAJOR >= 13:
         cdef object pref_loc = opts.preferred_location if opts is not None else None
         cdef object pref_type = opts.preferred_location_type if opts is not None else None
 
-        if pref_type is not None and pref_type not in _VALID_LOCATION_TYPES:
-            raise ValueError(
-                f"preferred_location_type must be one of {_VALID_LOCATION_TYPES!r} "
-                f"or None, got {pref_type!r}"
-            )
+        check_str_enum(pref_type, ManagedMemoryLocationType, allow_none=True)
 
         if pref_type is None:
             # Legacy behavior
diff --git a/cuda_core/cuda/core/_utils/validators.py b/cuda_core/cuda/core/_utils/validators.py
new file mode 100644
index 00000000000..dac3d1c788e
--- /dev/null
+++ b/cuda_core/cuda/core/_utils/validators.py
@@ -0,0 +1,35 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
+
+def format_or_list(values):
+    """Format an iterable of values as a human-readable ``A, B or C`` string.
+
+    Each value is passed through :func:`repr`. A single value is returned
+    as-is; two values are joined with ``" or "``; three or more use a
+    comma-separated list with ``" or "`` before the last item.
+    """
+    reprs = [repr(v) for v in values]
+    if len(reprs) <= 1:
+        return reprs[0] if reprs else ""
+    *head, tail = reprs
+    return ", ".join(head) + " or " + tail
+
+
+def check_str_enum(value, enum_class, *, allow_none=False):
+    """Raise ValueError if *value* is not a member of *enum_class*.
+
+    Derives the list of acceptable values from the enum itself so callers
+    do not need to maintain a parallel copy of the valid strings.
+
+    If *allow_none* is True, ``None`` is also accepted and included in the
+    error message when an invalid value is provided.
+    """
+    if allow_none and value is None:
+        return
+    if value not in {m.value for m in enum_class}:
+        valid = sorted(m.value for m in enum_class)
+        if allow_none:
+            valid = [None, *valid]
+        raise ValueError(f"{value!r} is not a valid {enum_class.__name__}. Must be {format_or_list(valid)}")
diff --git a/cuda_core/cuda/core/graph/_graph_node.pyx b/cuda_core/cuda/core/graph/_graph_node.pyx
index 077eff77ea0..0adf69ad4b4 100644
--- a/cuda_core/cuda/core/graph/_graph_node.pyx
+++ b/cuda_core/cuda/core/graph/_graph_node.pyx
@@ -65,6 +65,7 @@ import ctypes as ct
 import weakref
 
 from cuda.core.graph._adjacency_set_proxy import AdjacencySetProxy
+from cuda.core._utils.validators import check_str_enum
 from cuda.core._utils.cuda_utils import driver
 from cuda.core.typing import GraphMemoryType
 
@@ -787,6 +788,7 @@ cdef inline AllocNode GN_alloc(GraphNode self, size_t size, object device,
     alloc_params.poolProps.handleTypes = cydriver.CUmemAllocationHandleType.CU_MEM_HANDLE_TYPE_NONE
     alloc_params.bytesize = size
 
+    check_str_enum(memory_type_str, GraphMemoryType)
     if memory_type_str == "device":
         alloc_params.poolProps.allocType = cydriver.CUmemAllocationType.CU_MEM_ALLOCATION_TYPE_PINNED
         alloc_params.poolProps.location.type = cydriver.CUmemLocationType.CU_MEM_LOCATION_TYPE_DEVICE
@@ -802,9 +804,6 @@ cdef inline AllocNode GN_alloc(GraphNode self, size_t size, object device,
             alloc_params.poolProps.location.id = device_id
         ELSE:
             raise ValueError("memory_type='managed' requires CUDA 13.0 or later")
-    else:
-        raise ValueError(f"Invalid memory_type: {memory_type_str!r}. "
-                       "Must be 'device', 'host', or 'managed'.")
 
     if access_descs.size() > 0:
         alloc_params.accessDescs = access_descs.data()
diff --git a/cuda_core/cuda/core/utils/_program_cache/_keys.py b/cuda_core/cuda/core/utils/_program_cache/_keys.py
index 668a26257fb..e170bc18131 100644
--- a/cuda_core/cuda/core/utils/_program_cache/_keys.py
+++ b/cuda_core/cuda/core/utils/_program_cache/_keys.py
@@ -33,11 +33,12 @@
 from cuda.core._utils.cuda_utils import (
     nvrtc as _nvrtc,
 )
+from cuda.core._utils.validators import check_str_enum, format_or_list
+from cuda.core.typing import SourceCodeType
 
 # Bump when the key schema changes in a way that invalidates existing caches.
 _KEY_SCHEMA_VERSION = 2
 
-_VALID_CODE_TYPES = frozenset({"c++", "ptx", "nvvm"})
 _VALID_TARGET_TYPES = frozenset({"ptx", "cubin", "ltoir"})
 
 # code_type -> allowed target_type set, mirroring Program.compile's
@@ -752,10 +753,12 @@ def make_program_cache_key(
     # the lowercase form.
     code_type = code_type.lower() if isinstance(code_type, str) else code_type
     target_type = target_type.lower() if isinstance(target_type, str) else target_type
-    if code_type not in _VALID_CODE_TYPES:
-        raise ValueError(f"code_type={code_type!r} is not supported (must be one of {sorted(_VALID_CODE_TYPES)})")
+    check_str_enum(code_type, SourceCodeType)
     if target_type not in _VALID_TARGET_TYPES:
-        raise ValueError(f"target_type={target_type!r} is not supported (must be one of {sorted(_VALID_TARGET_TYPES)})")
+        raise ValueError(
+            f"target_type={target_type!r} is not supported by the program cache "
+            f"(must be {format_or_list(sorted(_VALID_TARGET_TYPES))})"
+        )
     supported_for_code = _SUPPORTED_TARGETS_BY_CODE_TYPE[code_type]
     if target_type not in supported_for_code:
         raise ValueError(
diff --git a/cuda_core/tests/graph/test_graph_definition.py b/cuda_core/tests/graph/test_graph_definition.py
index da78bea577f..50d4b4ac253 100644
--- a/cuda_core/tests/graph/test_graph_definition.py
+++ b/cuda_core/tests/graph/test_graph_definition.py
@@ -833,7 +833,7 @@ def test_alloc_free_chain(sample_graphdef):
 
 def test_alloc_memory_type_invalid(sample_graphdef):
     """Invalid memory type raises ValueError."""
-    with pytest.raises(ValueError, match="Invalid memory_type"):
+    with pytest.raises(ValueError, match="'invalid' is not a valid GraphMemoryType. Must be "):
         sample_graphdef.allocate(ALLOC_SIZE, memory_type="invalid")
 
 
diff --git a/cuda_core/tests/test_green_context.py b/cuda_core/tests/test_green_context.py
index effa7167371..999627a901a 100644
--- a/cuda_core/tests/test_green_context.py
+++ b/cuda_core/tests/test_green_context.py
@@ -252,7 +252,7 @@ def test_device_id_matches_source_multi_gpu(self):
         assert wq1.device.device_id == 1
 
     def test_invalid_scope_raises_at_construction(self):
-        with pytest.raises(ValueError, match="Unknown sharing_scope"):
+        with pytest.raises(ValueError, match="'bogus' is not a valid WorkqueueSharingScopeType. Must be "):
             WorkqueueResourceOptions(sharing_scope="bogus")
 
     def test_configure_concurrency_limit(self, wq_resource):
diff --git a/cuda_core/tests/test_memory.py b/cuda_core/tests/test_memory.py
index f53e8244328..5ef48919a50 100644
--- a/cuda_core/tests/test_memory.py
+++ b/cuda_core/tests/test_memory.py
@@ -1278,7 +1278,7 @@ def test_managed_memory_resource_preferred_location_validation(init_cuda):
     device.set_current()
 
     # Invalid preferred_location_type
-    with pytest.raises(ValueError, match="preferred_location_type must be one of"):
+    with pytest.raises(ValueError, match="'invalid' is not a valid ManagedMemoryLocationType. Must be "):
         ManagedMemoryResource(
             ManagedMemoryResourceOptions(
                 preferred_location_type="invalid",
diff --git a/cuda_core/tests/test_program_cache.py b/cuda_core/tests/test_program_cache.py
index 963ec1cc04b..ee393868640 100644
--- a/cuda_core/tests/test_program_cache.py
+++ b/cuda_core/tests/test_program_cache.py
@@ -317,7 +317,9 @@ def test_make_program_cache_key_rejects_extra_sources_outside_nvvm(code_type, co
 @pytest.mark.parametrize(
     "kwargs, exc_type, match",
     [
-        pytest.param({"code_type": "fortran"}, ValueError, "code_type", id="unknown_code_type"),
+        pytest.param(
+            {"code_type": "fortran"}, ValueError, "'fortran' is not a valid SourceCodeType", id="unknown_code_type"
+        ),
         pytest.param({"target_type": "exe"}, ValueError, "target_type", id="unknown_target_type"),
         pytest.param({"code": 12345}, TypeError, "code", id="non_str_bytes_code"),
         # Backend-specific target matrix -- Program.compile rejects these