Koopman Skew-Dilation RNN / report_bench_2026-09-01T102800.md
Mechanism confirmed, baseline not beaten
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 9, "verdict": "Built a local two-channel Koopman skew-dilation RNN using symmetric/skew parameterization and Cayley recurrence, evaluated end-to-end on the structurally matched actuated-pendulum dynamics track. The mechanism signature was confirmed on a trained model: Cayley relative norm error was 5.96e-7 while 20-step explicit-Euler norm ratio was 1.0036. However, the lift was significantly worse on test MSE than the tuned GRU baseline (delta_mean=+0.01630, permutation p=0.0081), so the promised training benefit did not transfer.", "metrics": { "baseline": "dynamics/rnn_small, tuned lr=0.01 and 12 epochs; 8-seed mean test MSE 0.0015438102, std 0.0006459184", "idea": "two-channel Cayley lift, best tested lr=0.01 and 12 epochs; 8-seed mean test MSE 0.0178441483, std 0.0168653353", "paired_delta": "idea - baseline = +0.0163003380 MSE; idea wins 0/8; permutation p=0.0081", "mechanism_signature": { "observed_cayley_relative_norm_error": 5.960464477539062e-07, "observed_euler_20_step_norm_ratio": 1.003595232963562, "confirmed": true } }, "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "epochs": 12 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 12 }, "mean": 0.004508913028985263, "std": 0.002062311810243919 }, { "cfg": { "lr": 0.003, "epochs": 12 }, "mean": 0.0030303404928417876, "std": 0.0014911902872439856 }, { "cfg": { "lr": 0.01, "epochs": 12 }, "mean": 0.0015438102273037657, "std": 0.0006459183721555687 } ], "full": { "per_seed": [ 0.0018195160664618015, 0.0012350792530924082, 0.0008721873164176941, 0.0027302715461701155, 0.0014946963638067245, 0.0011278137098997831, 0.0009376407833769917, 0.002133276779204607 ], "mean": 0.0015438102273037657, "std": 0.0006459183721555687 } }, "idea": { "best_cfg": { "lr": 0.01, "epochs": 12 }, "sweep": [ { "cfg": { "lr": 0.001, "epochs": 12 }, "mean": 0.0210957785602659, "std": 0.02216245868447279 }, { "cfg": { "lr": 0.003, "epochs": 12 }, "mean": 0.019453824614174664, "std": 0.01769262402163577 }, { "cfg": { "lr": 0.01, "epochs": 12 }, "mean": 0.01784414827125147, "std": 0.016865335320885074 } ], "per_seed": [ 0.010384509339928627, 0.014125057496130466, 0.052780672907829285, 0.03464017063379288, 0.008677596226334572, 0.00931015145033598, 0.0071034664288163185, 0.005731561686843634 ], "mean": 0.01784414827125147, "std": 0.016865335320885074 }, "comparison": { "delta_mean": 0.016300338043947704, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.008564993273466825, 0.012889978243038058, 0.05190848559141159, 0.03190989908762276, 0.007182899862527847, 0.008182337740436196, 0.006165825645439327, 0.0035982849076390266 ], "p_value": 0.0081, "mde": 0.014119949734496392, "mde_rel_pct": 914.6169318463842, "verdict": "idea worse (significant)", "system_worked": false }, "mechanism_signature": { "prediction": "Cayley preserves lifted norm; explicit Euler grows it", "observed_cayley_relative_norm_error": 5.960464477539062e-07, "observed_euler_20_step_norm_ratio": 1.003595232963562, "trained_test_mse": 0.01244353037327528, "confirmed": true }, "custom_track": null }, "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. The lift has a two-channel latent state (128 dimensions) versus the GRU's 64-dimensional hidden state, so parameter/capacity matching is imperfect; no long-horizon rollout, wall-clock/FLOP, larger channel counts, block-tridiagonal solver, or sequential-MNIST test was run.", "system_verdict": "partial", "practical_verdict": "harms", "mechanism_ok": 1, "system_judged": true }