calibrate: per-arm candidate counts and mild-arm offsets #82

Merged
weksa merged 3 commits from weksa/calibrate-arm-reports into main 2026-11-13 14:53:42 +00:00
Member

Follow-up calibrate-arm-counts-and-mild-offsets. weksa loop calibrate prints one candidate=LABEL line per candidate arm after the pooled one, and offset LANG=+x on each degraded= line (arm composite minus mean prompted composite, R included). CLI test pins the class-line order. Docs updated; map not yet re-checked against growth-scene.

🤖 Generated with Claude Code

Follow-up calibrate-arm-counts-and-mild-offsets. `weksa loop calibrate` prints one `candidate=LABEL` line per candidate arm after the pooled one, and `offset LANG=+x` on each `degraded=` line (arm composite minus mean prompted composite, R included). CLI test pins the class-line order. Docs updated; map not yet re-checked against growth-scene. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
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>
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>
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>
Author
Member

Soul pass (no cut_report, so no admitted verdict): first pass on 064483d found the unit test could not tell arms or languages apart (Medium, 8 of 13 mutants survived) and a vacuous CLI offset assertion (Low); fixed in a85d410. Re-attack: 6 of 8 killed; M7/M10 (CLI print layer) survived, so the degraded-line builder moved into the lib and is pinned by an asymmetric test (e94563b, mechanical, not re-attacked by Soul; I checked the suite and clippy). growth-scene re-run reproduced the map: h3 45/60, h4 40/60 all; new h3 50/60, h4 0/60; mild offsets new en -0.0041, ja +0.0134 (mild-1) / +0.0054 (mild-2), old en -0.0721. Last commit is a map wording fix.

Soul pass (no cut_report, so no admitted verdict): first pass on 064483d found the unit test could not tell arms or languages apart (Medium, 8 of 13 mutants survived) and a vacuous CLI offset assertion (Low); fixed in a85d410. Re-attack: 6 of 8 killed; M7/M10 (CLI print layer) survived, so the degraded-line builder moved into the lib and is pinned by an asymmetric test (e94563b, mechanical, not re-attacked by Soul; I checked the suite and clippy). growth-scene re-run reproduced the map: h3 45/60, h4 40/60 all; new h3 50/60, h4 0/60; mild offsets new en -0.0041, ja +0.0134 (mild-1) / +0.0054 (mild-2), old en -0.0721. Last commit is a map wording fix.
weksa merged commit 36297f8d97 into main 2026-11-13 14:53:42 +00:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
GameCult/Weksa!82
No description provided.