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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "The cross-ratio recurrent mechanism was evaluated against a matched standard recurrent baseline on the registered dynamics track. The trained-model mechanism signature was confirmed, but the idea had worse mean test MSE and no significant paired improvement, so worked=false.", "metrics": { "baseline": "Mean test MSE 0.004535878775641322; best configuration lr=0.003, epochs=25, weight_decay=0.0001.", "idea": "Mean test MSE 0.004785479104612023; paired delta +0.0002496003289707005; 3/8 wins; permutation p=0.209." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_cross_ratio_bench.py", "files": [ "stage2_cross_ratio_bench.py", "bench_report.json" ], "limitations": "Only the registered built-in dynamics track was tested. Long-horizon rollout error, Lyapunov exponents, homogeneous projective handling, and additional recurrent architectures 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": 0.003, "epochs": 25, "weight_decay": 0.0001 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 25, "weight_decay": 0.0 }, "mean": 0.006130807567387819 }, { "cfg": { "lr": 0.001, "epochs": 25, "weight_decay": 0.0001 }, "mean": 0.006120537058450282 }, { "cfg": { "lr": 0.003, "epochs": 25, "weight_decay": 0.0 }, "mean": 0.0046020153677091 }, { "cfg": { "lr": 0.003, "epochs": 25, "weight_decay": 0.0001 }, "mean": 0.0045645206701010466 }, { "cfg": { "lr": 0.01, "epochs": 25, "weight_decay": 0.0 }, "mean": 0.004882440727669746 }, { "cfg": { "lr": 0.01, "epochs": 25, "weight_decay": 0.0001 }, "mean": 0.004718217242043465 } ], "full": { "mean": 0.004535878775641322, "std": 0.0004771483174727427, "per_seed": [ 0.004597439430654049, 0.0038385987281799316, 0.004727847874164581, 0.005094196647405624, 0.004155295435339212, 0.00398264080286026, 0.00463221687823534, 0.005258794408291578 ], "n": 8 } }, "idea": { "mean": 0.004785479104612023, "std": 0.00089246428498962, "per_seed": [ 0.004405152052640915, 0.0037003301549702883, 0.00573378661647439, 0.005794585682451725, 0.0041740648448467255, 0.0035881984513252974, 0.004965356085449457, 0.005922358948737383 ], "n": 8 }, "comparison": { "delta_mean": 0.0002496003289707005, "idea_wins": 3, "n_pairs": 8, "per_seed_diffs": [ -0.000192287378013134, -0.00013826857320964336, 0.0010059387423098087, 0.0007003890350461006, 1.876940951301304e-05, -0.00039444235153496265, 0.00033313920721411705, 0.0006635645404458046 ], "p_value": 0.209, "mde": 0.0004201483148548282, "mde_rel_pct": 9.262776534309474, "verdict": "no significant win", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.003, "epochs": 25, "weight_decay": 0.0001 }, "result": { "mean": 0.006392467068508267, "std": 0.0020169420755724718, "per_seed": [ 0.005178917199373245, 0.01132385153323412, 0.006093441043049097, 0.0059410021640360355, 0.006832372397184372, 0.006475373636931181, 0.004938049241900444, 0.004356729332357645 ], "n": 8 } }, { "cfg": { "lr": 0.001, "epochs": 25, "weight_decay": 0.0 }, "result": { "mean": 0.022261007456108928, "std": 0.02048209560797128, "per_seed": [ 0.014691504649817944, 0.015150383114814758, 0.07583310455083847, 0.014610606245696545, 0.010224367491900921, 0.017106497660279274, 0.02021658793091774, 0.01025500800460577 ], "n": 8 } }, { "cfg": { "lr": 0.01, "epochs": 25, "weight_decay": 0.0001 }, "result": { "mean": 0.004785479104612023, "std": 0.00089246428498962, "per_seed": [ 0.004405152052640915, 0.0037003301549702883, 0.00573378661647439, 0.005794585682451725, 0.0041740648448467255, 0.0035881984513252974, 0.004965356085449457, 0.005922358948737383 ], "n": 8 } } ], "structural_match": "stability/control dynamics", "mechanism_signature": { "prediction": "trained model plaquette residual near zero", "observed_mean_residual": 3.852884162824921e-07, "observed_singular_fraction": 0.0, "predicted_float32_scale": "~1e-6", "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "inconclusive", "mechanism_ok": 1, "system_judged": true }