Absence-of-Seam as a Finding

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Two callers sharing a counter lose an update, but existing tests create a separate counter per caller. They stay green without covering the real race.

Mechanism

Retain reproducible inputs, scheduling and shared-state identity. Check whether the test boundary exercises the actual two-caller chain. If not, report the missing coordination seam and why current tests cannot lock the bug down. Propose a minimal controlled two-caller harness checking final count. Keep meaningful existing checks and report incomplete coverage until the gap is filled. Structural changes follow authorized scope.

Bad example

Add an isolated-counter unit test and claim permanent race coverage without mentioning shared-state differences.

Good example

Explain that per-caller counters cannot reproduce this shared race. Design a minimal controlled test with two updates to one counter: old behavior should yield 1 versus expected 2. Retain repro and missing-boundary records, marking regression coverage incomplete until tested.

Why the change matters

A boundary excluding the trigger cannot gain race coverage from more green tests. Reporting the missing seam identifies the architecture obstructing durable verification.

Observable expectation

A teaching schedule lets both callers read 0 then write 1, ending at 1; correct atomic updates end at 2. Ensure the harness shares the same counter and exercises conflicting interleaving rather than a mock returning the expected count. This is a design, not an execution record.

Limits

A missing ideal seam is no excuse to abandon meaningful checks; preserve the minimized repro. Process, storage and network conditions may matter, and an in-memory harness covers only simulated interleavings. Avoid exposing excessive internals merely for testing.