Adversarial Code Review for Passing Tests
When tests pass but implementation may have hidden flaws, conduct adversarial review by: (1) identify sentinel values or default parameters that conflate different states (e.g., decline_rate=0.0 as both "not set" and "valid input"), (2) check boundary conditions for unjustified constraints (e.g., b_factor < 1 when b=1 is mathematically valid), (3) distinguish between user intent and implementation convenience. Fix MAJOR findings, add regression tests for edge cases, then re-run full suite before committing.