chore(eval): integrate Retrieval Eval v2 and close Phase 1 - #215

Merged

mrsibe merged 19 commits into

mainfrom

chore/eval-v2-closeout

Oct 1, 2026

Commits on Sep 30, 2026

fix(eval): run the harness on Windows and keep the baseline POSIX
mrsibe
committed
25e1ada
feat(eval): separate candidateK from contextK, and run the production config
mrsibe
committed
a444a86
feat(eval): hit rate, MAP, and per-query-type metrics
mrsibe
committed
54c892f
feat(eval): derive the similarity threshold on validation, report on test
mrsibe
committed
a2ccf05
feat(eval): adopt on the first metric with headroom, not a saturated one
mrsibe
committed
a52b9b7
feat(eval): measure the context window, and name what still needs a model
mrsibe
committed
b77257a
feat(eval): bounded parameter sweep with a trade-off dashboard
mrsibe
committed
27fdda2
fix(eval): refuse a context window wider than the retrieval depth
mrsibe
committed
aee2f89
feat(eval): unanswerable questions, and an explicit split manifest
mrsibe
committed
d452463
feat(eval): nine cross-lingual questions, and the corpus stops being saturated
mrsibe
committed
3f60d7c
fix(retrieval): trace the dense channel's threshold for hybrid too
mrsibe
committed
b1bdb0b
fix(eval): select the strategy on validation, and stop calling candidates "passages"
mrsibe
committed
84e9e0a
feat(eval): a confusion cluster, and the corpus gets its first hard negatives
mrsibe
committed
20a8c75
feat(eval): a second confusion cluster, and the corpus reaches 53 chunks
mrsibe
committed
3ffee33
feat(eval): dense score diagnostics, and the answer is not a threshold
mrsibe
committed
08fa7c1
fix(eval): convert a score margin with ×2, not with the absolute affine transform
mrsibe
committed
036a95a
feat(eval): per-question paired deltas, and hybrid's gain is five questions
mrsibe
committed
8a99107
feat(eval): fifteen near-miss unanswerable questions, and the last eval PR
mrsibe
committed
d368832

Commits on Oct 1, 2026

chore(eval): integrate retrieval v2 and close Phase 1 scope
mrsibe
committed
a850f9f

Navigation Menu

chore(eval): integrate Retrieval Eval v2 and close Phase 1 - #215

chore(eval): integrate Retrieval Eval v2 and close Phase 1#215
mrsibe merged 19 commits into
mainfrom
chore/eval-v2-closeout

Commits on Sep 30, 2026

fix(eval): run the harness on Windows and keep the baseline POSIX

feat(eval): separate candidateK from contextK, and run the production config

feat(eval): hit rate, MAP, and per-query-type metrics

feat(eval): derive the similarity threshold on validation, report on test

feat(eval): adopt on the first metric with headroom, not a saturated one

feat(eval): measure the context window, and name what still needs a model

feat(eval): bounded parameter sweep with a trade-off dashboard

fix(eval): refuse a context window wider than the retrieval depth

feat(eval): unanswerable questions, and an explicit split manifest

feat(eval): nine cross-lingual questions, and the corpus stops being saturated

fix(retrieval): trace the dense channel's threshold for hybrid too

fix(eval): select the strategy on validation, and stop calling candidates "passages"

feat(eval): a confusion cluster, and the corpus gets its first hard negatives

feat(eval): a second confusion cluster, and the corpus reaches 53 chunks

feat(eval): dense score diagnostics, and the answer is not a threshold

fix(eval): convert a score margin with ×2, not with the absolute affine transform

feat(eval): per-question paired deltas, and hybrid's gain is five questions

feat(eval): fifteen near-miss unanswerable questions, and the last eval PR

Commits on Oct 1, 2026

chore(eval): integrate retrieval v2 and close Phase 1 scope