fix: quick-dev review classification misses data-loss findings as defer #2138
Description
Description
The bad_spec classification in step-04-review.md does not cover findings where data is silently dropped or never reaches its destination within the story's domain. This causes such findings to be misclassified as defer (pre-existing) even when they fall within the initiative's scope.
Problem
In a real-world scenario, an edge case hunter correctly identified that a CreditCheck builder never mapped a field (learningPopulation) that its parallel sibling module already mapped. The finding was classified as defer (pre-existing) because the diff didn't touch that line — but the missing mapping was within the initiative's goals and was a blocking data-loss bug.
The current bad_spec definition only covers "caused by the change, including direct deviations from spec." It doesn't account for findings where data is silently dropped within the story's domain, even if the code predates the diff.
Proposed Fix
Extend the bad_spec definition in step-04-review.md to include data-loss findings within the story's domain, preventing them from being misclassified as defer.
Context
Discovered through a postmortem after a quick-dev workflow shipped a spec that missed a field mapping. The edge case hunter found it, but classification killed the finding.