Diagnose an API test that fails only after another case

Understand the likely causes

• Common causes include shared mutable nested fixtures, a fake left installed beyond its intended scope, a reused identifier, cached subject context, or an injected clock not restored after a case. A genuine product defect is also possible. Failure order alone does not distinguish those causes, so preserve competing explanations until a minimal reproduction separates them.

Preserve the decisive order and boundary

Copy the smallest relevant synthetic fixture into temporary storage and identify what each case creates, mutates, replaces, and owns. Compare the failing pair in the original order with each case in isolation. Do not point the test at a configured database or real account to make the result more realistic. Inventory global dictionaries, nested lists, class attributes, cached clients, environment readers, and dependency replacements. If the test claims authentication coverage but directly injects a subject, correct the evidence label while continuing the isolation diagnosis.

Test the nested-copy hypothesis locally

A shallow copy creates a new outer dictionary while retaining references to nested objects. The example demonstrates that specific mechanism with public classroom identifiers. First it confirms the shared-list symptom, then it constructs independent state and verifies the repair. This is a controlled reproduction of fixture leakage, not proof that the application has the same defect. If the real failing fixture is immutable or freshly constructed, reject this hypothesis and inspect replacement lifetime or cached context instead of forcing the symptom into a familiar explanation.

from copy import deepcopy
base = {"entries": ["book-1"]}
shallow = dict(base)
shallow["entries"].append("book-2")
assert base["entries"] == ["book-1", "book-2"]

first = deepcopy({"entries": ["book-1"]})
second = deepcopy({"entries": ["book-1"]})
first["entries"].append("book-2")
assert second["entries"] == ["book-1"]
print("shared-list cause reproduced; independent fixture checked")

shared-list cause reproduced; independent fixture checked

Repair ownership rather than adding delay

Create fresh state at the appropriate scope, restore replacements through the test framework’s supported cleanup mechanism, and inject deterministic clocks and identifiers. If a shared resource is intentional, define its owner and reset rule explicitly. Avoid broad cleanup that deletes resources owned by other cases, and avoid arbitrary sleeps or test-order pinning as the final fix. Check whether the correction masks a missing authorization or validation branch. Keep the reproduction small enough that a reviewer can explain exactly why the original order leaked state.

Keep it from recurring

• Keep fixture ownership explicit, prohibit accidental mutable globals, review nested structures and replacement lifetimes, and retain the smallest order-dependent regression. Record seeds and actual generated inputs for randomized cases. A passing rerun is supporting evidence only when the cause and corrected boundary are understood.