Six scripted pi sub-agent runs (3 tool regimes × 2 models) rehearsed the
exercise; deterministic gate = solution test + approved file. Key results:
mandated CLI triples cost on the strong model (4:06 vs 0:39, PASS both);
weak model fails only under the mandate (A+Q), passes when free or optional.
Drift documented with code: scorer.updateSelection left inside the extracted
method (A+Q, C+Q). Harness in tools/rehearse; summary in
rehearsal-findings.md; per-run evidence in rehearsal-runs/; cheat sheet in
rehearse-with-agent.md corrected (lambda-typed parameters unsupported).
- Samman learning hour: 5/5/10/25/5 outline; pairs stop at extracted
method taking the lambda as a parameter
- exercise/java: pre-factored variant of the kata (Q1/Q4 dead branch
stripped, dates as constant), approval test red/flaky by design
- exercise/solution: collectExpectedWeather(LongFunction<Boolean>) +
no-arg overload; deterministic approval test (all-sunny lambda)
- rehearse-with-agent.md: rehearsal setup with qwen + java-refactoring-mcp
Your Name