From 9783a338ce4452da986e05eeae0f0ac6db4ab45a Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 15 Nov 2026 15:07:49 +0000 Subject: [PATCH] Gre24 spec: fidelity-first criterion, three-list reporting, one-factor R4 attribution Co-Authored-By: Claude Opus 5.5 --- docs/claude/gre24-reproduction-spec-2026-11-15.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/docs/claude/gre24-reproduction-spec-2026-11-15.md b/docs/claude/gre24-reproduction-spec-2026-11-15.md index 8f6e4b709..17d488d06 100644 --- a/docs/claude/gre24-reproduction-spec-2026-11-15.md +++ b/docs/claude/gre24-reproduction-spec-2026-11-15.md @@ -35,12 +35,13 @@ Source: Greif, Lechner, Meyer, Composites Part A 185 (2024) 108340, author PDF f - Data: v4 (10027027) if downloadable at a reasonable size, otherwise v5 (624 parts, flagged as a version deviation). - Target: t* = t_fill V-dot/V if V-dot is available, otherwise t_fill/t_set, flagged. - **R3 (MeltAstra, run).** The paper protocol: 5 random 315/314 splits (seed fixed and recorded), naive + GBRT + FNN, per-geometry RRMSE over **all nodes**, and min/median/max over the 5 splits. Budget: up to 40 worker CPU-h and up to 2 GPU-h tonight; per-part guards as before; failures kept. -- **Reproduction acceptance (fixed now, before any result).** "Reproduced on the paper's protocol" means: +- **Fidelity is the core criterion (TRON clarification).** The reproduction is judged by how closely method and protocol match the paper. The published numbers are a *diagnostic reference for deviations*, not a validation gate. Landing inside the ranges below does not by itself prove a faithful reproduction, and landing outside does not refute one. The report keeps three lists separate: (1) paper facts, (2) our implementation choices for ambiguous items, (3) items that cannot be reproduced (missing or unavailable). +- **Numeric reference ranges (fixed now, before any result, diagnostic only):** - GBRT and FNN fill-time median RRMSE (primary normaliser) in [0.028, 0.040], around the published 0.034; - the naive model is within about 25 % of GBRT's median (the paper's "ML only slightly better than naive"); - the max stays in the same regime (0.2 to 0.3). - Anything outside these is reported as a deviation with its likely cause (data version, RRMSE normaliser, target scaling), not tuned away. -- **R4 (after R3, possibly tomorrow).** The same method on our grouped 617/209 protocol and S2 query domain. Compare against R2-G24 and the S2 B0, both at the paper's per-geometry RRMSE and at our component-mean relative L2. This answers directly how much of the R2/B0 gap was the Dijkstra approximation. +- **R4 (after R3, possibly tomorrow).** The same method on our grouped 617/209 protocol and S2 query domain. Compare against R2-G24 and the S2 B0, both at the paper's per-geometry RRMSE and at our component-mean relative L2. Attribution must change one factor at a time, because R4 otherwise changes distance, features and protocol together. Step (a): naive V*_f with eikonal D_f versus naive V*_f with our tet-edge Dijkstra D_f, with the same V*_f measure, roster, query domain and metric. That isolates the distance construction. Step (b): naive versus GBRT on the eikonal features (the value of the features and the model). Step (c): R2-G24 versus the Gre24 GBRT. No step's difference is attributed to a factor it did not isolate. ## 3. Note on a result that matters Gre24 reports that its learned models are **only slightly better than the naive V*_f model** on fill time. In our S2, R2 (7.30 %) is well ahead of our Dijkstra B0 (10.40 %). If the faithful eikonal naive model closes most of that gap, then our B0 was mostly a weak distance construction, and the remaining public fill-time problem is smaller than S2 suggested.