{ "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.005, "K": 15 }, "sweep": [ { "cfg": { "lr": 0.001, "K": 3 }, "mean": 0.016172005562111735 }, { "cfg": { "lr": 0.003, "K": 3 }, "mean": 0.003774263139348477 }, { "cfg": { "lr": 0.005, "K": 3 }, "mean": 0.0017951192276086658 }, { "cfg": { "lr": 0.001, "K": 7 }, "mean": 0.015029519563540816 }, { "cfg": { "lr": 0.003, "K": 7 }, "mean": 0.002934314659796655 }, { "cfg": { "lr": 0.005, "K": 7 }, "mean": 0.0014960496919229627 }, { "cfg": { "lr": 0.001, "K": 15 }, "mean": 0.014895078958943486 }, { "cfg": { "lr": 0.003, "K": 15 }, "mean": 0.0028665231075137854 }, { "cfg": { "lr": 0.005, "K": 15 }, "mean": 0.0013684573350474238 } ], "full": { "mean": 0.0011932104389416054, "std": 0.000203301503237927, "per_seed": [ 0.0015044162282720208, 0.0011529525509104133, 0.0014893548795953393, 0.001327105681411922, 0.001068409881554544, 0.0009947591461241245, 0.0010155801428481936, 0.0009931050008162856 ], "n": 8 } }, "idea": { "mean": 0.0023188923223642632, "std": 0.0018174311282403578, "per_seed": [ 0.006926196161657572, 0.0018025230383500457, 0.0029087376315146685, 0.0010936673497781157, 0.0015447858022525907, 0.0014355253661051393, 0.0016264201840385795, 0.0012132830452173948 ], "n": 8 }, "comparison": { "delta_mean": 0.0011256818834226578, "idea_wins": 1, "n_pairs": 8, "per_seed_diffs": [ 0.005421779933385551, 0.0006495704874396324, 0.0014193827519193292, -0.00023343833163380623, 0.0004763759206980467, 0.00044076621998101473, 0.0006108400411903858, 0.00022017804440110922 ], "p_value": 0.02285, "mde": 0.0015021645326744831, "mde_rel_pct": 125.89267438918188, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "structural_match": "controlled pendulum rollout with recurrent fixed-point state updates", "shared_architecture": "same inp/state/head parameters and K evaluations; only baseline vs Fejer update differs", "idea_grid": [ { "cfg": { "lr": 0.001, "K": 3 }, "mean": 0.013947936473414302, "per_seed": [ 0.02009131945669651, 0.008569814264774323, 0.04837782308459282, 0.007685891818255186, 0.005474410951137543, 0.006128633860498667, 0.0073469290509819984, 0.007908669300377369 ] }, { "cfg": { "lr": 0.003, "K": 3 }, "mean": 0.004136359391850419, "per_seed": [ 0.002831853926181793, 0.0031103442888706923, 0.016510576009750366, 0.0014188818167895079, 0.0029145670123398304, 0.0021216198801994324, 0.0022474690340459347, 0.0019355631666257977 ] }, { "cfg": { "lr": 0.005, "K": 3 }, "mean": 0.0023188923223642632, "per_seed": [ 0.006926196161657572, 0.0018025230383500457, 0.0029087376315146685, 0.0010936673497781157, 0.0015447858022525907, 0.0014355253661051393, 0.0016264201840385795, 0.0012132830452173948 ] }, { "cfg": { "lr": 0.001, "K": 7 }, "mean": 0.06781748728826642, "per_seed": [ 0.09825821965932846, 0.060682933777570724, 0.1300727277994156, 0.10404244810342789, 0.02876441925764084, 0.03181176632642746, 0.04144725203514099, 0.04746013134717941 ] }, { "cfg": { "lr": 0.003, "K": 7 }, "mean": 0.022045168676413596, "per_seed": [ 0.03464265167713165, 0.02418467029929161, 0.060446545481681824, 0.0186628308147192, 0.011261790990829468, 0.007364522200077772, 0.007728805299848318, 0.01206953264772892 ] }, { "cfg": { "lr": 0.005, "K": 7 }, "mean": 0.028678134316578507, "per_seed": [ 0.027619337663054466, 0.009325508028268814, 0.04537428915500641, 0.10992163419723511, 0.011131452396512032, 0.00906284898519516, 0.009421450085937977, 0.007568554021418095 ] }, { "cfg": { "lr": 0.001, "K": 15 }, "mean": 0.45379000529646873, "per_seed": [ 0.5679315328598022, 0.6005325317382812, 0.31230661273002625, 0.5303172469139099, 0.5320272445678711, 0.35211363434791565, 0.2027447521686554, 0.5323464870452881 ] }, { "cfg": { "lr": 0.003, "K": 15 }, "mean": 0.31799950916320086, "per_seed": [ 0.25116634368896484, 0.3567662239074707, 0.34843429923057556, 0.7776817083358765, 0.16537804901599884, 0.31844574213027954, 0.09809694439172745, 0.22802676260471344 ] }, { "cfg": { "lr": 0.005, "K": 15 }, "mean": 0.2506284995470196, "per_seed": [ 0.42000678181648254, 0.17470629513263702, 0.2908547818660736, 0.1604318916797638, 0.34779635071754456, 0.018573837354779243, 0.14079605042934418, 0.45186200737953186 ] } ], "mechanism_signature": { "prediction": "Fejer residual contracts approximately as 1/(K+1) under resolvent/nonexpansive assumptions", "K": 7, "predicted_ratio": 0.125, "observed_trained_ratio": 1.12704236805439, "identity_error_on_trained_model": 13.323629379272461, "confirmed": false } } }