diff --git a/collector/pyproject.toml b/collector/pyproject.toml
index a88a165..490ffe1 100644
--- a/collector/pyproject.toml
+++ b/collector/pyproject.toml
@@ -33,6 +33,10 @@ addopts = "-q --doctest-modules"
 # Coverage column reports against, and it is deliberately narrower than
 # jq_local_code_lines, which counts every tracked source file in the repo.
 source = ["jq_collector"]
+# Both arms of every branch, not just every line: an `if` whose false side
+# never runs is 100% line-covered and still untested. fail_under below then
+# holds lines and branches to the same floor.
+branch = true
 
 [tool.coverage.report]
 # The package is fully covered; this keeps it that way. pytest-cov reads it, so
diff --git a/collector/tests/test_coverage.py b/collector/tests/test_coverage.py
index c56437f..8dcbbd7 100644
--- a/collector/tests/test_coverage.py
+++ b/collector/tests/test_coverage.py
@@ -58,6 +58,21 @@ def test_the_newest_default_branch_artifact_wins_not_the_newest_overall(make_cli
     assert client.coverage_artifact("o/r", "main") == 2
 
 
+def test_an_older_artifact_listed_after_a_newer_one_does_not_win(make_client):
+    """Newest wins by created_at, not by where it sits in the listing."""
+    client = make_client(
+        {
+            ARTIFACTS: {
+                "artifacts": [
+                    artifact(2, "main", "2026-08-27T04:07:41Z"),  # newest on main
+                    artifact(1, "main", "2026-08-26T04:07:41Z"),  # older, listed after
+                ]
+            }
+        }
+    )
+    assert client.coverage_artifact("o/r", "main") == 2
+
+
 def test_expired_and_unrelated_artifacts_are_ignored(make_client):
     client = make_client(
         {
diff --git a/collector/tests/test_main.py b/collector/tests/test_main.py
index 5b1ea7b..b120970 100644
--- a/collector/tests/test_main.py
+++ b/collector/tests/test_main.py
@@ -312,6 +312,23 @@ def test_main_warns_when_there_is_no_token(monkeypatch, caplog):
     assert "no GITHUB_TOKEN" in caplog.text
 
 
+def test_main_does_not_warn_when_a_token_is_set(monkeypatch, caplog):
+    """The warning is for the deployment that needs it; a set token is silent."""
+    monkeypatch.setattr(entry, "_refresh_remote", lambda *_: None)
+    monkeypatch.setattr(entry, "_tick_local", lambda *_: None)
+    monkeypatch.setattr(entry, "start_http_server", lambda _port: None)
+    monkeypatch.setattr(entry.REGISTRY, "register", lambda _c: None)
+    monkeypatch.setattr(entry.signal, "signal", lambda _sig, handler: handler(_sig, None))
+
+    cfg = Config()
+    object.__setattr__(cfg, "token", "ghp_test")
+    monkeypatch.setattr(entry, "Config", lambda: cfg)
+
+    entry.main()
+
+    assert "no GITHUB_TOKEN" not in caplog.text
+
+
 def test_main_starts_a_loop_per_source_and_shuts_them_down(monkeypatch):
     started: list[str] = []
     real_thread = threading.Thread
diff --git a/collector/tests/test_workflows.py b/collector/tests/test_workflows.py
index c1542cd..6d8c442 100644
--- a/collector/tests/test_workflows.py
+++ b/collector/tests/test_workflows.py
@@ -83,6 +83,23 @@ def test_latest_is_by_completion_not_feed_order(make_client):
     assert c.latest_runs("o/r", "main")[0]["conclusion"] == "failure"
 
 
+def test_an_older_run_later_in_the_feed_does_not_replace_a_newer_one(make_client):
+    """The mirror of the case above: when the feed does happen to put the newer
+    run first, the older one after it must not win just by coming last."""
+    c = make_client(
+        {
+            WFS: {"workflows": [workflow(1, "CI")]},
+            RUNS: {
+                "workflow_runs": [
+                    run(1, "CI", "failure", "2026-08-01T05:39:00Z"),
+                    run(1, "CI", "success", "2026-08-01T04:09:00Z"),
+                ]
+            },
+        }
+    )
+    assert c.latest_runs("o/r", "main")[0]["conclusion"] == "failure"
+
+
 def test_workflow_missing_from_the_feed_is_fetched_directly(make_client):
     """The feed's first page is dominated by frequent workflows. cvxgrp/simulator
     has 2406 runs on main and only 6 of its 23 workflows appear in the first 100;
@@ -145,6 +162,27 @@ def test_two_active_workflows_sharing_a_name_get_distinct_labels(make_client):
     assert len(names) == len(set(names)), f"labels collide: {names}"
 
 
+def test_a_shared_name_with_no_path_to_fall_back_on_keeps_its_name(make_client):
+    """Only a workflow that has a path can be relabelled by it. One without
+    keeps its display name rather than being blanked, and the other still
+    moves to its path, so the labels stay distinct."""
+    pathless = workflow(1, "CI")
+    del pathless["path"]
+    c = make_client(
+        {
+            WFS: {"workflows": [pathless, workflow(2, "CI", path=".github/workflows/b.yml")]},
+            RUNS: {
+                "workflow_runs": [
+                    run(1, "CI", "success", "2026-08-01T00:00:00Z"),
+                    run(2, "CI", "failure", "2026-08-01T00:00:00Z"),
+                ]
+            },
+        }
+    )
+    got = {r["_name"]: r["conclusion"] for r in c.latest_runs("o/r", "main")}
+    assert got == {"CI": "success", ".github/workflows/b.yml": "failure"}
+
+
 def test_cancelled_run_falls_back_to_the_last_real_verdict(make_client):
     """Stopping a job by hand, or a concurrency group killing it when the next
     push lands, is not a failure. The workflow keeps its last conclusive run."""