Chern-Gap Monitor for Finite-Horizon Collapse / report_bench_2026-09-02T164639.md
Mechanism confirmed, baseline not beaten
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 10, "verdict": "Implemented the Chern-gap barrier on the registered dynamics track with identical rnn_small systems, matched learning-rate sweeps, and eight paired seeds. The trained-model signature confirmed constant rounded Chern sector while the measured gap remained nonzero, but test MSE was marginally worse and non-significant (delta_mean=+0.0000755, p=0.82525).", "metrics": { "baseline": "Best lr=0.003; 8-seed test MSE mean=1.5021538734436035, std=0.09324438560832148.", "idea": "Best lr=0.003; 8-seed test MSE mean=1.5022293776273727, std=0.09287950268012642." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_chern_gap.py > bench_output.json", "files": [ "bench_chern_gap.py", "bench_output.json" ], "limitations": "Only the built-in dynamics track was tested. The intervention was a gap barrier rather than early stopping; no longer horizon, larger model, or reflection-symmetry-breaking study was run.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.003 }, "sweep": [ { "cfg": { "lr": 0.001 }, "mean": 1.596962183713913 }, { "cfg": { "lr": 0.003 }, "mean": 1.552410066127777 }, { "cfg": { "lr": 0.01 }, "mean": 1.5703844726085663 } ], "full": { "mean": 1.5021538734436035, "std": 0.09324438560832148, "per_seed": [ 1.4778438806533813, 1.4308412075042725, 1.5888422727584839, 1.7121129035949707, 1.4590684175491333, 1.4697362184524536, 1.4667506217956543, 1.4120354652404785 ], "n": 8 } }, "idea": { "mean": 1.5022293776273727, "std": 0.09287950268012642, "per_seed": [ 1.4786245822906494, 1.4313784837722778, 1.5867565870285034, 1.7121129035949707, 1.459558367729187, 1.4695805311203003, 1.4680730104446411, 1.4117505550384521 ], "n": 8 }, "comparison": { "delta_mean": 7.550418376922607e-05, "idea_wins": 3, "n_pairs": 8, "per_seed_diffs": [ 0.0007807016372680664, 0.0005372762680053711, -0.0020856857299804688, 0.0, 0.0004899501800537109, -0.0001556873321533203, 0.0013223886489868164, -0.0002849102020263672 ], "p_value": 0.82525, "mde": 0.0008536970299377886, "mde_rel_pct": 0.05683153004696756, "verdict": "no measurable effect", "system_worked": false }, "idea_sweep": [ { "cfg": { "lr": 0.003 }, "mean": 1.5022293776273727 }, { "cfg": { "lr": 0.001 }, "mean": 1.5371521562337875 }, { "cfg": { "lr": 0.01 }, "mean": 1.5104802548885345 } ], "mechanism_signature": { "probe": "trained RNN predictions under three fixed phase perturbations", "predicted": "Rounded Chern remains constant while gap is nonzero; sector changes require gap closing.", "observed_transition_count": 0, "min_checkpoint_gap": 0.005794507451355457, "seed0_checkpoints": [ { "epoch": 1, "gap": 0.0251762717962265, "rounded_chern": 0 }, { "epoch": 3, "gap": 0.058473870158195496, "rounded_chern": 0 }, { "epoch": 5, "gap": 0.054124489426612854, "rounded_chern": 0 }, { "epoch": 7, "gap": 0.01788416877388954, "rounded_chern": 0 }, { "epoch": 9, "gap": 0.015780657529830933, "rounded_chern": 0 }, { "epoch": 10, "gap": 0.005794507451355457, "rounded_chern": 0 } ], "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "no_effect", "mechanism_ok": 1, "system_judged": true }