c2e74fe2fd
The result that should have been visible for weeks. Measured on the new holdout_dev lane (500 real held-out cases) + sealed test: - real capability 0% (train_sample's 4 correct generalize to 0/500 held-out) - resolve_pooled (the composition substrate) commits 17.4% WRONG on held-out - NO structural gate (step/pool/op) separates its 2 correct from 87 wrong -> the composer cannot tell right readings from wrong; gating it is overfitting - the only wrong=0 policy on real GSM8K is refusal (current serving 0/0/500) Real lift needs a verifiable reading mechanism, not composer tuning. Non-serving (docs). |
||
|---|---|---|
| .. | ||
| composition-capability-scope.md | ||
| composition-wall-execution-plan-2026-06-03.md | ||
| comprehension-primitive-inventory.md | ||
| extraction-richness-audit-2026-06-03.md | ||
| gsm8k-lift-program-strategy-2026-06-04.md | ||
| question-layer-gap-survey.md | ||
| real-gsm8k-capability-measurement-2026-06-04.md | ||
| solver-operation-coverage.md | ||
| wiring-promotion-gate-brief-2026-06-03.md | ||