Cyclic Lie-Bracket Residual Block / report_bench_2026-08-31T222908.md

✓✓ Beats tuned baseline

Raw ⬇ ZIP

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

(универсальный)

Ответ агента:

{ "worked": true, "confidence": 7, "verdict": "Implemented the cyclic Lie-bracket residual block on the structurally matched dynamics track using the same GRU encoder and paired 8-seed protocol. The tuned cyclic model achieved mean test MSE 0.0001866451 versus 0.0003950244 for the tuned baseline, winning on all 8 seeds; paired delta was -0.0002083793 with permutation p=0.0081. The trained-model order gap was nonzero, but its observed small-step exponent was 2.9885 rather than the predicted 2.0, so the mechanism prediction was not quantitatively confirmed.", "metrics": { "baseline": "dynamics/rnn_small, 12 epochs, lr=0.006: mean test MSE 0.0003950244, std 0.0001292329; 8 per-seed results [0.00040816, 0.00056587, 0.00031052, 0.00062471, 0.00027074, 0.00039440, 0.00022947, 0.00035633]. Baseline lr sweep [0.001, 0.003, 0.006] means [0.00136616, 0.00073085, 0.00039502].", "idea": "cyclic centered two-field GRU residual, 12 epochs, lr=0.006, step=0.1: mean test MSE 0.0001866451, std 0.0000630393; all 8 paired seeds improved. Idea step sweep [0.1, 0.15, 0.22] means [0.00018665, 0.00019070, 0.00020175]. Paired delta=-0.0002083793, permutation p=0.0081, 8/8 wins.", "mechanism_signature": "Trained-model order gap at step 0.1 was 0.0122054. Gaps at steps [0.01, 0.02, 0.04] were [1.3433e-05, 1.0711e-04, 8.4608e-04], giving observed exponent 2.9885 versus predicted 2.0; confirmed=false." }, "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "step": 0.0 }, "sweep": [ { "cfg": { "lr": 0.001, "step": 0.0 }, "mean": 0.0013661633856827393 }, { "cfg": { "lr": 0.003, "step": 0.0 }, "mean": 0.0007308490021387115 }, { "cfg": { "lr": 0.006, "step": 0.0 }, "mean": 0.00039502444451500196 } ], "full": { "cfg": { "lr": 0.006, "step": 0.0 }, "mean": 0.00039502444451500196, "std": 0.00012923294272245648, "per_seed": [ 0.0004081559309270233, 0.0005658670561388135, 0.00031052311533130705, 0.0006247127312235534, 0.00027073538512922823, 0.00039440052933059633, 0.00022947481193114072, 0.00035632599610835314 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.006, "step": 0.1 }, "sweep": [ { "cfg": { "lr": 0.006, "step": 0.1 }, "mean": 0.00018664514300326118 }, { "cfg": { "lr": 0.006, "step": 0.15 }, "mean": 0.0001907046962514869 }, { "cfg": { "lr": 0.006, "step": 0.22 }, "mean": 0.00020175300687697018 } ], "full": { "mean": 0.00018664514300326118, "std": 6.30393270484372e-05, "per_seed": [ 0.0002500905538909137, 0.00021891914366278797, 0.0001631042396184057, 0.0002553996164351702, 9.480195876676589e-05, 0.00010218997340416536, 0.00015309224545489997, 0.00025556341279298067 ], "n": 8 } }, "comparison": { "delta_mean": -0.00020837930151174078, "idea_wins": 8, "n_pairs": 8, "per_seed_diffs": [ -0.00015806537703610957, -0.0003469479124760255, -0.00014741887571290135, -0.00036931311478838325, -0.00017593342636246234, -0.00029221055592643097, -7.638256647624075e-05, -0.00010076258331537247 ], "p_value": 0.0081, "mde": 9.39821994106377e-05, "mde_rel_pct": 23.791489543394196, "verdict": "idea better (significant)", "system_worked": true }, "mechanism_signature": { "order_gap_mean": 0.012205412611365318, "steps": [ 0.009999999776482582, 0.019999999552965164, 0.03999999910593033 ], "gaps": [ 1.3433276762953028e-05, 0.00010710919741541147, 0.0008460786193609238 ], "predicted_exponent": 2.0, "observed_exponent": 2.9884542718353977, "confirmed": false } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_cyclic.py", "files": [ "bench_cyclic.py", "bench_report.json" ], "limitations": "Only the built-in dynamics track was tested, with 1000 training samples, 300 test samples, 12 epochs, and a small rnn_small model. No vision, sequence, corruption robustness, FLOP timing, gradient-variance analysis, commuting-field control, or larger-scale dynamics evaluation was run. The measured mechanism exponent did not match the predicted quadratic scaling.", "system_verdict": "worked", "practical_verdict": "helps", "mechanism_ok": 0, "system_judged": true }