Bidirectional Saturation-Aware Trust Region / report_bench_2026-09-02T170937.md

Failed on benchmark

Raw ⬇ ZIP

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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Implemented the bidirectional saturation-aware trust-region intervention for the structurally matched tabular optimizer track, using the same MLP and Adam learning-rate search space for baseline and idea. Across 8 paired seeds, the idea was significantly worse: test MSE 92.2100 versus 8.5679 for baseline Adam, paired delta +83.6422, permutation p=0.0081. The trained-model mechanism signature was not confirmed because q remained 1.0 and the radius expanded without entering the predicted contraction regime.", "metrics": { "baseline": "Adam, best lr=0.01: mean test MSE 8.5679, std 0.9045; sweep means lr=0.001: 25.6016, lr=0.003: 13.8608, lr=0.01: 8.3528.", "idea": "Adaptive Adam-proposal trust region, best tested lr=0.001 and r0=0.02: mean test MSE 92.2100, std 10.3835. Mean q=1.0, clipping fraction=1.0, mean scale=0.1577, radius 0.02 to 0.0844." }, "bench_report": { "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 25.60164165496826 }, { "cfg": { "lr": 0.003 }, "mean": 13.860755920410156 }, { "cfg": { "lr": 0.01 }, "mean": 8.352817177772522 } ], "full": { "mean": 8.567867696285248, "std": 0.9045233203943077, "per_seed": [ 7.95448637008667, 9.628232955932617, 7.726639747619629, 8.101909637451172, 7.126259803771973, 8.949170112609863, 9.36670970916748, 9.689533233642578 ], "n": 8 } }, "idea": { "mean": 92.21004962921143, "std": 10.383480498694608, "per_seed": [ 99.29408264160156, 93.70222473144531, 98.78123474121094, 72.97736358642578, 81.31901550292969, 94.73509216308594, 88.57316589355469, 108.2982177734375 ], "n": 8 }, "comparison": { "delta_mean": 83.64218193292618, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 91.33959627151489, 84.0739917755127, 91.05459499359131, 64.87545394897461, 74.19275569915723, 85.78592205047607, 79.2064561843872, 98.60868453979492 ], "p_value": 0.0081, "mde": 8.961902038617074, "mde_rel_pct": 104.598977905584, "verdict": "idea worse (significant)", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.001, "r0": 0.02 }, "mean": 91.18872833251953 }, { "cfg": { "lr": 0.003, "r0": 0.02 }, "mean": 91.1887264251709 }, { "cfg": { "lr": 0.01, "r0": 0.02 }, "mean": 91.18872833251953 } ], "mechanism_signature": { "mean_q": 1.0, "clip_fraction": 1.0, "radius_final": 0.08441391633993081, "radius_initial": 0.02, "mean_scale": 0.15767846046228018, "mean_proposal_norm": 0.2890951458546632, "prediction": "sustained saturation expands radius and later unsaturation contracts it", "observed_radius_increase": true, "observed_clipping_fraction": 1.0, "observed_unsaturation": false, "observed_radius_contraction": false, "confirmed": false, "confirmation_reason": "Radius expanded, but q remained 1.0 and no post-saturation contraction was observed." } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py", "files": [ "bench_experiment.py", "bench_report.json" ], "limitations": "Only the structurally matched tabular track and mlp_tiny model were tested. The adaptive implementation used a single global radius, did not test vision, sequence, or dynamics, and did not evaluate curvature-based r_max or alternative optimizers. The experiment nevertheless followed the required 8 paired seeds, baseline sweep, shared learning-rate union, and permutation test.", "system_verdict": "failed", "practical_verdict": "harms", "mechanism_ok": 0, "system_judged": true }