Lipschitz-Free Metric Pooling / report_bench_2026-09-04T134741.md

Failed on benchmark

Raw ⬇ ZIP

Стенд-проверка (stage-2) · промт оператора:

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Implemented end-to-end Lipschitz-Free Metric Pooling on the registered sequence forecasting track with a bounded finite probe bank and matched mean-pooling baseline. Across 8 paired seeds, the idea was significantly worse: MSE 1.028305 versus 1.010802, paired delta +0.017503, permutation p=0.0163. The trained-model signature found zero Lipschitz violations but did not confirm positive geometric association.", "metrics": { "baseline": "Mean pooling, best lr=0.003: MSE 1.010802±0.092335 across 8 seeds.", "idea": "K=8 bounded probe pooling, best lr=0.006: MSE 1.028305±0.091821 across 8 seeds; 1/8 paired wins." }, "bench_report": { "bench_version": 1, "track": "sequence", "model": "transformer_tiny_pooling_mvp", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003, "epochs": 12, "batch": 128 }, "sweep": [ { "cfg": { "lr": 0.0015, "epochs": 12, "batch": 128 }, "mean": 1.030422881245613 }, { "cfg": { "lr": 0.003, "epochs": 12, "batch": 128 }, "mean": 1.0044431388378143 }, { "cfg": { "lr": 0.006, "epochs": 12, "batch": 128 }, "mean": 1.0057344734668732 } ], "full": { "mean": 1.0108022689819336, "std": 0.09233493114632767, "per_seed": [ 1.135109543800354, 0.8252689242362976, 1.0580546855926514, 0.9993394017219543, 1.0295724868774414, 1.0314586162567139, 1.09066903546156, 0.9169454574584961 ], "n": 8 } }, "idea": { "mean": 1.0283051505684853, "std": 0.09182080119670853, "per_seed": [ 1.1634190082550049, 0.8420974612236023, 1.050006628036499, 1.0190545320510864, 1.0613138675689697, 1.0465973615646362, 1.1009020805358887, 0.9430502653121948 ], "n": 8, "selected_cfg": { "lr": 0.006, "epochs": 12, "batch": 128 } }, "comparison": { "delta_mean": 0.017502881586551666, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ 0.02830946445465088, 0.016828536987304688, -0.008048057556152344, 0.01971513032913208, 0.03174138069152832, 0.015138745307922363, 0.010233044624328613, 0.02610480785369873 ], "p_value": 0.0163, "mde": 0.010529986179047253, "mde_rel_pct": 1.041745403287733, "verdict": "idea worse (significant)", "system_worked": false }, "structural_match": "sequence-level window forecasting with token pooling", "parameter_counts": { "baseline": 321, "idea": 337 }, "selection_seeds": [ 0, 1, 2, 3 ], "epochs": 12, "batch": 128, "idea_sweep": [ { "cfg": { "lr": 0.0015, "epochs": 12, "batch": 128 }, "mean": 1.091338686645031, "std": 0.09665476674907689, "per_seed": [ 1.2205655574798584, 0.9110400080680847, 1.1126540899276733, 1.0519979000091553, 1.12910795211792, 1.1222410202026367, 1.1943774223327637, 0.9887255430221558 ] }, { "cfg": { "lr": 0.003, "epochs": 12, "batch": 128 }, "mean": 1.075037308037281, "std": 0.09951274952114578, "per_seed": [ 1.2106199264526367, 0.8868147134780884, 1.1060532331466675, 1.03895902633667, 1.1129647493362427, 1.103129506111145, 1.173069715499878, 0.9686875939369202 ] }, { "cfg": { "lr": 0.006, "epochs": 12, "batch": 128 }, "mean": 1.0283051505684853, "std": 0.09182080119670853, "per_seed": [ 1.1634190082550049, 0.8420974612236023, 1.050006628036499, 1.0190545320510864, 1.0613138675689697, 1.0465973615646362, 1.1009020805358887, 0.9430502653121948 ] } ], "mechanism_signature": { "quantity": "trained probe D_hat versus independent 1D Wasserstein-1", "predicted": "D_hat <= W1 for every pair; positive geometric association", "observed_mean_D_hat": 0.027728637424297632, "observed_mean_W1": 0.31382585523523404, "mean_ratio_D_hat_over_W1": 0.09807194932021986, "pearson_correlation": 0.12651039453842908, "violation_fraction": 0.0, "confirmed": false } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json" ], "limitations": "Only the registered sequence forecasting track was tested. The implementation used scalar sequence values, fixed equal weights, no learned router, no explicit pairwise Lipschitz penalty, and no vision, dynamics, retrieval, point-cloud, or removal-robustness evaluation.", "system_verdict": "failed", "practical_verdict": "harms", "mechanism_ok": 0, "system_judged": true }