calibrate: per-arm candidate counts and mild-arm offsets #82
2026-11-13
calibrate: degraded lines built in the lib and pinned by the arm-and-language test
Soul arms s1 re-check: CLI-only mutants M7 and M10 (offsets zipped against the wrong language or label) survived. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
calibrate: tests that tell arms and languages apart; map notes the harness now prints them
Soul arms s1 finding: unit test could not distinguish arms, languages or the R-included mean. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
calibrate: per-arm candidate counts, mild-arm composite offsets, line order pinned
Trials gains candidate_arms (pooled candidate is their sum) and degraded_offset (arm composite minus the prompted mean, per language). The CLI test pins the order of the class lines and the per-arm candidate lines. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>