code-quorumlisted

# Code quorum Ordinary code review finds style issues and obvious bugs reliably. What it misses is the class that matters: **the interaction that only fails under a specific input, at a specific scale, on the second deploy.** This adapter seats lenses that hunt those, with an evidence bar that rejects taste dressed as correctness. --- ## When to run it | Run it | Skip it | |---|---| | Irreversible or hard-to-reverse changes — migrations, data deletion, auth, money movement | Formatting, renames, comments | | Changes to a subsystem you did not write | Changes with strong existing test coverage and a clean diff | | Before a merge nobody else will review | Prototypes and spikes | | After the second failed fix attempt on the same bug | | That last row is the highest-value trigger. **Two failed fixes means the model of the problem is wrong**, and more attempts from the same model produce more wrong fixes. A quorum re-derives the model from evidence. --- ## The lenses ### 1. Correctness adversary > *"Show me an input that produces a wrong result."* Hunts: boundary values (zero, one, empty, max), off-by-one, null and undefined through the happy path, unicode and encoding, timezone and DST boundaries, floating-point money arithmetic, integer overflow, sort stability assumptions, silent type coercion. Must produce **a concrete input**, not a category. "Could have an off-by-one" refutes; "with an empty array, line 42 indexes `[0]` and throws" does not. ### 2. Failure advers