Land the five rulings and force the insufficient_evidence line (#508) - #519

Merged

pengfei-threemoonslab merged 10 commits into

mainfrom

claude/issue-508-ie-calibration

Sep 5, 2026

Commits on Sep 3, 2026

Land the five rulings, both calibration re-runs, and the three cases that force the evidence-gap line (#508)

pengfei-threemoonslab
and
claude
committed
b46dd48

Commits on Sep 4, 2026

Address PR #519 review: staging erased link changes, equal bytes are not identity, and the recorded model was the announced one

pengfei-threemoonslab
and
claude
committed
c07033a
Widen the answer-key check past the one file it knew about

pengfei-threemoonslab
and
claude
committed
ec25527
Make the answer-free host a script instead of a thing to remember

pengfei-threemoonslab
and
claude
committed
bcb301b
Run the first corpus round: 96 labels, and two blockers that stop the tag (#508)

pengfei-threemoonslab
and
claude
committed
c6523ce
Guide: a binding is a capability you can name; insufficient_evidence is a signal, not a step (#520)

pengfei-threemoonslab
and
claude
committed
51964a2
Corpus round 2 on the corrected guide: kappa 0.8048 on 47 pairs, one case without a label (#508, #520)

pengfei-threemoonslab
and
claude
committed
6f2954b
Record the six adjudications, and the third blocker they expose (#508, #520)

pengfei-threemoonslab
and
claude
committed
39b60e2
Retire `insufficient_evidence` as a target, and shorten four cells to the material that exists (#520, #508)

pengfei-threemoonslab
and
claude
committed
0c0be62

Commits on Sep 5, 2026

Address PR #519 review: a guide seam that split two raters, and three guards that were narrower than their own reasons (#508, #520)

pengfei-threemoonslab
and
claude
committed
cbe2324