Trajectory-Learned Actuator-Aware Funnel Network / report_bench_2026-08-31T091800.md
Beats tuned baseline
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": true, "confidence": 8, "verdict": "The actuator-aware funnel intervention significantly improved test MSE on the registered dynamics track using the same rnn_small architecture. It won all 8 paired seeds, with delta_mean=-0.0006042214 and permutation p=0.0081. The trained-model mechanism signature confirmed bounded authority, although the authority bound is partly enforced by tanh.", "metrics": { "baseline": "best lr=0.01; mean test MSE 0.0006267142; std 0.0001539544", "idea": "best lr=0.01; mean test MSE 0.00002249282; std 0.00000416654", "delta_mean": -0.0006042213872206048, "p_value": 0.0081, "idea_wins": "8/8" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_funnel.py", "files": [ "bench_funnel.py", "bench_report.json" ], "limitations": "Only the registered built-in dynamics track with 400 training samples, 400 test samples, and 15 epochs was tested. The intervention is a one-step bounded correction rather than a full multi-step rollout; other actuator limits, funnel quantiles, horizons, disturbances, and out-of-distribution states were not tested.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 0.005253629118669778 }, { "cfg": { "lr": 0.003 }, "mean": 0.001200278929900378 }, { "cfg": { "lr": 0.01 }, "mean": 0.0006579469481948763 } ], "full": { "mean": 0.0006267142089200206, "std": 0.0001539544331865635, "per_seed": [ 0.0006342190317809582, 0.0005908075836487114, 0.0004556644707918167, 0.0009510967065580189, 0.000620594946667552, 0.00040738406823948026, 0.0006597545580007136, 0.0006941923056729138 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.01 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 3.181707052135607e-05 }, { "cfg": { "lr": 0.003 }, "mean": 2.614848335724673e-05 }, { "cfg": { "lr": 0.01 }, "mean": 2.301071633168841e-05 } ], "mean": 2.24928216994158e-05, "std": 4.166535651612339e-06, "per_seed": [ 2.433888403174933e-05, 2.1393136194092225e-05, 2.2563392121810466e-05, 2.3747454179101627e-05, 2.724600926740095e-05, 2.3439619326381944e-05, 2.4891782231861725e-05, 1.2322296242928132e-05 ], "n": 8 }, "comparison": { "delta_mean": -0.0006042213872206048, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.0006098801477492088, -0.0005694144474546192, -0.00043310107867000625, -0.0009273492523789173, -0.000593348937400151, -0.0003839444489130983, -0.0006348627757688519, -0.0006818700094299857 ], "p_value": 0.0081, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "predicted_kmax": 1.5, "requested_authority_mean": 0.08706140890717506, "observed_authority_mean": 0.08706140890717506, "authority_ratio_mean": 0.99999998849756, "funnel_violation_rate_mean": 0.0, "confirmed": true } }, "system_verdict": "worked", "practical_verdict": "helps", "mechanism_ok": 1, "system_judged": true }