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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Complete Interval Abstraction Training was implemented as an interval-loss modification of the shared rnn_small dynamics model. The tuned baseline achieved mean test MSE 0.00355935 versus 0.00360583 for the idea; paired delta was +0.00004647 with permutation p=0.81695, so there was no significant win. The trained-model mechanism signature was not confirmed, with containment 0.7031 and mean violation 0.01052.", "metrics": { "baseline": "rnn_small, lr=0.006, epochs=10; full 8-seed mean MSE 0.0035593539651017636.", "idea": "rnn_small with interval loss, lr=0.006, epochs=10; full 8-seed mean MSE 0.0036058283294551075; 3/8 paired wins." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py", "files": [ "bench_experiment.py", "bench_report.json", "bench_run.txt" ], "limitations": "Only the built-in actuated-pendulum dynamics task was tested. Multi-step abstract reachability, policy optimization, learned residual calibration, and alternative monotone architectures were not evaluated.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "epochs": 10 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 10 }, "mean": 0.02098797250073403 }, { "cfg": { "lr": 0.003, "epochs": 10 }, "mean": 0.003764481545658782 }, { "cfg": { "lr": 0.006, "epochs": 10 }, "mean": 0.002816903230268508 } ], "full": { "mean": 0.0035593539651017636, "std": 0.001430033924501602, "per_seed": [ 0.003789895214140415, 0.0012226301478222013, 0.004499303642660379, 0.001755783916451037, 0.005420389119535685, 0.0051945955492556095, 0.0037620163056999445, 0.0028302178252488375 ], "n": 8 } }, "idea": { "mean": 0.0036058283294551075, "std": 0.0012201842547257958, "per_seed": [ 0.004097327124327421, 0.0013652343768626451, 0.0035981065593659878, 0.0025071303825825453, 0.0055099669843912125, 0.0048635597340762615, 0.0037585990503430367, 0.0031467024236917496 ], "n": 8 }, "comparison": { "delta_mean": 4.6474364353343844e-05, "idea_wins": 3, "n_pairs": 8, "per_seed_diffs": [ 0.000307431910187006, 0.0001426042290404439, -0.0009011970832943916, 0.0007513464661315084, 8.957786485552788e-05, -0.000331035815179348, -3.4172553569078445e-06, 0.0003164845984429121 ], "p_value": 0.81695, "mde": 0.00041109428784617027, "mde_rel_pct": 11.549688282671736, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "predicted": "smaller h should reduce interval width while residual inflation preserves containment", "observed_containment_mean": 0.7031250298023224, "observed_width_mean": 0.1384880207479, "observed_violation_mean": 0.010517213333514519, "confirmed": false }, "sanity_check": { "h": [ 0.2, 0.1, 0.05, 0.025 ], "containment": [ 1.0, 1.0, 1.0, 1.0 ], "mean_width": [ 0.2894333333333333, 0.18980000000000002, 0.14003333333333334, 0.11466666666666667 ], "all_contained": true, "width_decreases": true }, "idea_config_sweep": [ { "cfg": { "lr": 0.006, "epochs": 10 }, "mean": 0.0036058283294551075 }, { "cfg": { "lr": 0.001, "epochs": 10 }, "mean": 0.023580193694215268 }, { "cfg": { "lr": 0.003, "epochs": 10 }, "mean": 0.004334091427153908 } ], "custom_track": null }, "system_verdict": "failed", "practical_verdict": "no_effect", "mechanism_ok": 0, "system_judged": true }