Nonlinear Noise-Tightening Drift / report_bench_2026-08-31T095529.md

Failed on benchmark

Raw ⬇ ZIP

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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 10, "verdict": "The cubic restoring drift was implemented end-to-end in a GRU predictor on the structurally matched controlled-pendulum dynamics track. The best idea was significantly worse than the swept baseline, and its trained-model contraction signature did not confirm the predicted lower bound.", "metrics": { "baseline": "lr=0.003; mean test MSE 7.923793088338016e-08 over 8 seeds.", "idea": "beta=0.05, lr=0.01; mean test MSE 1.2224308409969353e-07 over 8 seeds; paired delta +4.3005153216313374e-08; p=0.0081; 0/8 wins.", "mechanism_signature": "Predicted lower bound 0.981; observed mean effective restoring rate 0.03263185149990022; fraction above bound 0.0; confirmed=false." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json", "nonlinear_drift.py", "run_experiment.py" ], "limitations": "Used 1200 training examples, 400 test examples, and 12 epochs for runtime. Only the built-in controlled-pendulum dynamics track was tested; no multidimensional stochastic stationary-covariance benchmark was run.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003 }, "sweep": [ { "lr": 0.001, "mean": 1.5209165837859473e-07 }, { "lr": 0.003, "mean": 7.913097554990145e-08 }, { "lr": 0.01, "mean": 8.186839473012242e-08 } ], "full_mean": 7.923793088338016e-08 }, "idea": { "config": { "beta": 0.05, "lr": 0.01 }, "mean": 1.2224308409969353e-07 }, "comparison": { "delta_mean": 4.3005153216313374e-08, "p_value": 0.0081, "idea_wins": 0, "n_pairs": 8, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "predicted_lower_bound": 0.981, "observed_mean_rate": 0.03263185149990022, "observed_fraction_above_lambda": 0.0, "confirmed": false } }, "system_verdict": "failed", "practical_verdict": "harms", "mechanism_ok": 0, "system_judged": true }