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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Implemented an end-to-end tabular MLP variant using a shared rank-16 trained-model Jacobian subspace and Woodbury-style gradient preconditioning, with identical architecture, data, epochs, and hyperparameter grid as the Adam baseline. Mean test MSE improved from 9.774816 to 8.963684, but the paired permutation p-value was 0.10015, so the mandatory harness verdict is no significant win. The honest trained-model signature found only 21.3% mean held-out Jacobian energy captured by the rank-16 subspace, so the proposed transfer mechanism was not confirmed.", "metrics": { "baseline": "Tabular/Friedman#1 mlp_tiny, Adam, best lr=0.006 and weight_decay=0.0001; 8-seed mean test MSE 9.774816, std 1.146524.", "idea": "Shared rank-16 Jacobian Woodbury preconditioning, best lr=0.006 and weight_decay=0.0001; 8-seed mean test MSE 8.963684, std 1.218498.", "delta_mean": "-0.811133 MSE (-8.30% relative)", "idea_wins": "6/8 paired seeds", "permutation_p_value": 0.10015, "mechanism_signature_mean_captured_fraction": 0.21334, "mechanism_signature_confirmed": false }, "bench_report": { "bench_version": 1, "track": "tabular", "model": "mlp_tiny", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0001 }, "mean": 16.405594 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 13.136068 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "mean": 9.729963 } ], "full": { "mean": 9.774816, "std": 1.146524, "per_seed": [ 10.155755, 11.133987, 8.66899, 8.961121, 7.614502, 10.594099, 10.266081, 10.803994 ], "n": 8 } }, "idea": { "mean": 8.963684, "std": 1.218498, "per_seed": [ 10.10267, 11.417047, 9.143009, 8.528031, 7.162959, 8.287326, 8.166311, 8.902117 ], "n": 8 }, "comparison": { "delta_mean": -0.811133, "idea_wins": 6, "n_pairs": 8, "per_seed_diffs": [ -0.053085, 0.283059, 0.474019, -0.433089, -0.451544, -2.306773, -2.09977, -1.901877 ], "p_value": 0.10015, "verdict": "no significant win", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0001 }, "mean": 17.797435 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0001 }, "mean": 13.299632 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0001 }, "mean": 9.79769 } ], "mechanism_signature": { "type": "trained_model_jacobian_low_rank", "rank": 16, "predicted_vs_observed_mean_ratio": 0.21334, "confirmed": false } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py", "files": [ "bench_experiment.py", "bench_report.json" ], "limitations": "Only the tabular/Friedman#1 track and mlp_tiny architecture were tested. The intervention was an optimizer-level Woodbury preconditioner rather than full Laplace evidence or repeated neural-network hyperparameter marginal-likelihood evaluation; wall-clock and memory savings for candidate evidence evaluation were not measured. The low-rank subspace used an exact small calibration SVD rather than randomized Lanczos, and only rank 16 was evaluated.", "system_verdict": "failed", "practical_verdict": "inconclusive", "mechanism_ok": 0, "system_judged": true }