Continuation Maps for Training-Mode Transitions / report_bench_2026-09-01T132905.md
Mechanism confirmed, baseline not beaten
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 10, "verdict": "Implemented continuation-based late-window oscillation control on the structurally matched dynamics track using the shared rnn_small architecture, equal learning-rate sweeps, and 8 paired seeds. The mechanism signature was confirmed: mean oscillation feature fell from 0.5832 to 0.1645, a 71.8% reduction, but test MSE worsened significantly from 0.002912 to 0.262946 (paired delta +0.260034, permutation p=0.0081). The diagnostic effect is real, but the proposed intervention did not improve the standard task metric.", "metrics": { "baseline": "Best fixed-SGD lr=0.01; test MSE mean 0.0029119684, std 0.0019628704", "idea": "Best continuation lr=0.01; test MSE mean 0.2629458737, std 0.0624234057; oscillation feature 0.1644519 versus baseline 0.5831796; relative reduction 71.80%; final lr mean 0.00029296875", "paired_delta": 0.26003390522964764, "permutation_p_value": 0.0081, "comparison_verdict": "idea worse (significant)" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json" ], "limitations": "Only the built-in dynamics track was tested because stability/control is structurally matched there. The correction was a short noisy local controller rather than a full multi-point boundary map with checkpoint reuse; schedule-following and batch-size/noise continuation studies were not run.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "track_match": "stability/control -> dynamics", "baseline": { "best_cfg": { "lr": 0.01 }, "grid": [ { "lr": 0.001 }, { "lr": 0.003 }, { "lr": 0.01 } ], "full_mean": 0.0029119684331817552, "full_std": 0.0019628704033910396, "per_seed": [ 0.0013499484630301595, 0.002197778085246682, 0.0036032795906066895, 0.00771956006065011, 0.0013576406054198742, 0.0016320815775543451, 0.002993914531543851, 0.0024415445514023304 ] }, "idea": { "selected_cfg": { "lr": 0.01 }, "grid": [ { "lr": 0.001 }, { "lr": 0.003 }, { "lr": 0.01 } ], "mean": 0.2629458736628294, "std": 0.06242340567924847, "per_seed": [ 0.3094339966773987, 0.3548787236213684, 0.20706547796726227, 0.3201036751270294, 0.2438417375087738, 0.22770699858665466, 0.15347930788993835, 0.2870570719242096 ] }, "comparison": { "delta_mean": 0.26003390522964764, "idea_wins": 0, "n_pairs": 8, "p_value": 0.0081, "verdict": "idea worse (significant)" }, "mechanism_signature": { "prediction": "Continuation should reduce late-window loss oscillation while remaining on the stable side.", "baseline_feature_mean": 0.583179637221284, "idea_feature_mean": 0.16445192880727005, "relative_feature_reduction": 0.7180081088023521, "observed_final_lr_mean": 0.00029296875000000004, "n_trained_models": 16, "confirmed": true }, "baseline_grid": [ { "lr": 0.001 }, { "lr": 0.003 }, { "lr": 0.01 } ], "idea_grid": [ { "lr": 0.001 }, { "lr": 0.003 }, { "lr": 0.01 } ], "paired_seeds": [ 0, 1, 2, 3, 4, 5, 6, 7 ] }, "system_verdict": "partial", "practical_verdict": "harms", "mechanism_ok": 1, "system_judged": true }