research-tree: correct item-1 g sign reading

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
ClaudeandClaude Opus 5.5 committed 2026-11-18 06:29:15 +00:00
1 parent 9e2b687bd3
commit 3c05a466c1
1 file changed
+1 -1
+1 -1
View File
@@ -34,7 +34,7 @@ Rules:
| P3 | O1 with size-matched K = 8 partitions | — | — | PRUNED (eligibility 23 %; lesson: control cut area as a covariate, do not match it) |
Log (newest first):
- 2026-11-18: **Phase 3 item 1 (data sufficiency) COMPLETE: UNIDENTIFIED_OR_OVERFIT_SIGNS** (pre-registered rule; validation and test). Data-quantity effect within 25-100 % not identified (not proof of no benefit or of overfitting). Update budget with complete schedule lowered error (test g_50 -.0040, CI [-.0079, -.0003]; g_100 -.0027, CI crosses 0; validation g_100 -.0029, CI [-.0053, -.0005]); budget and schedule changed jointly. One fold / seed; CIs exclude training randomness. Postfit was recovered once after an infrastructure GPU-busy stop (TRON-launched, labelled). **Consequences:** Track D trigger not met (pretraining stays parked); a longer-schedule H2 arm is a new candidate after seed 1 (MoldAstra spec). Track A H2 training running; postfit pre-score PASS (983bba99). Seed 1 entry in preparation (dynamic dispatch on free GPUs).
- 2026-11-18: **Phase 3 item 1 (data sufficiency) COMPLETE: UNIDENTIFIED_OR_OVERFIT_SIGNS** (pre-registered rule; validation and test). **Corrected reading (my first reading had the g sign backwards; MeltAstra correction):** g = error(U) − error(2U), so the negative g values mean the 2U schedule had *higher* error (test 50 %: .0688 → .0728; 100 %: .0668 → .0696; validation 100 %: .0584 → .0613), i.e. overfitting signs, not a gain from longer training. s_2U = (error50 − error100)/log(n100/n50) > 0 means more data helps: validation +.0033, CI [+.0011, +.0055]; test +.0032, CI crosses 0. One fold / seed; budget and schedule changed jointly; the rule's classification stands. Track D trigger not formally met; the longer-schedule candidate is withdrawn (direction was wrong); overfitting control is the better-aimed candidate after seed 1.
- 2026-11-18: **S2 last-fill diagnostic (descriptive, post hoc, no gate; MoldAstra spec 05344bac; result 98519a6c).** 552 / 204. Full-node peak class A / H2: near the true peak 153 / 129; inside the true last-1 % tail but away from the peak 350 / 350; *outside the true tail (far)* 49 / 71 (component-weighted 9.4 % / 16.4 %); exact tie at another true peak 0 / 2. The γ > 2ε certificate holds for 0 / 552 in every combination (errors are too large relative to peak margins for the sufficient condition, so it is uninformative). **Reading:** H2's last-fill loss is mostly *more far-from-tail picks* plus fewer near-peak picks, not exact-tie flips; my hypothesis (b) (metric fragility) is at most a partial explanation. For about 63 % of parts both models land inside the true late region but not at its peak, so the set-based endpoint matters. This supports item c (location-aware objective) as designed; nothing is decided by this diagnostic.
- 2026-11-18: **Phase 3 item 1 (data sufficiency, v2) running.** Design revised from e24fff7c: six fresh fits (25 % / 50 % at 200 epochs; 50 % and 100 % each at U = 32k and 2U = 64k steps with their own complete schedules), 100 % 200-epoch reused from S1; U vs 2U = joint budget+schedule change. Pre-run review PASS (MeltAssist cbd67318); cap 120,000 GPU-s. S2 published (2338dfe9) with independent terminal audit PASS (d4999e4a); folds 1-4 alone: field Δ -.00495, CI [-.00765, -.00248].
- 2026-11-18: **H2 CONFIRMED on field (pre-registered pooled rule), last fill WORSE.** 552 / 204 test parts, pre-score review PASS 9c163702; postfit launched by TRON as a labelled one-time exception (MeltAstra unreachable). Field H2 .06579 vs A .07083: Δ -.00504, CI [-.00754, -.00283] (about 7 % relative); vs B13-E1b Δ -.00391, CI [-.00631, -.00164]. τ +.0035, CI [+.0010, +.0062]. **Last fill H2 .4566 vs A .3767: Δ +.080, CI [+.007, +.156] (worse).** Boundary: the pooled rule was registered after fold 0 was seen, so fold 0 is not independent; folds 1-4 alone to be reported descriptively. Reading: the learned representation is the lever for the field (better ordering, lower error), but it puts the last-filled point in the wrong place more often than A. Phase 3 is triggered (data sufficiency → seed 1 → c last-fill in H2 → b → seed 2 / ensemble); c is now the priority question.