Crucible

Say what in a diff is wrong, unproven or unearned, with evidence a reader can rerun.

Origin: Google's diff-scoped mutation testing (Petrović et al. 2021: one mutant per changed line took useful mutants from 15% to 89%) and Shi et al. 2014 (pruning a suite by killed mutants loses no fault detection; by coverage, up to 20.5%).

Findings are about correctness, the spec and the user's written rules, not taste: a reviewer asked to find gaps "will usually report some" even when the work is sound. The spec is the ticket's body, and it is optional; without one, steps that need it say so and the rest run.

Steps