# Evaluation cases

Use `../evaluation-cases.md` as the exercise material. The expected outcomes below are an answer key. Keep the answer key hidden from an agent during an evaluation run.

| Case | Expected finding | Reason |
| --- | --- | --- |
| A: new gate per press | Find ineffective admission ownership | Every press gets independent pending state |
| B: release only after successful await | Find blocked retry after rejection | The reset is skipped when the task rejects |
| C: one retained gate with finally | No concurrency/retry defect established by the snippet | Still check the call site and the scope of tests |
| D: disabled is false but retained handler guard exists | Do not claim duplicate requests from the prop alone | Report missing busy feedback separately if the requirement includes it |

Evaluate supported findings and false positives separately. Keep runtime uncertainties visible. A perfect result on these four prepared cases does not establish general performance.