Mistake Clinic

Module-Specific Mistake Radar

• Treating fluent confidence as verification: demand an independent example or source.
• Letting generated tests define the requirement: derive expectations first.
• Using eval to run a proposal: keep the lab schema declarative and bounded.
• Claiming fixture scores are model accuracy: label authored data and coverage.
• Counting copied solutions as independent learning: preserve an unaided transfer attempt.

Practice Mistake Checks

Choose one mistake, predict a concrete failing result and add the smallest case that distinguishes it from correct behavior. Repair only after recording the observation. Re-run the affected checks and explain what the test still cannot prove. Keep the original failure in your private evidence record.