Interval-Certified Equilibrium Layer / report_bench_2026-09-02T161537.md
Failed on benchmark
Стенд-проверка (stage-2) · промт оператора:
(универсальный)
Ответ агента:
{ "worked": false, "confidence": 10, "verdict": "Implemented an interval-certified implicit tanh recurrent layer and evaluated it on the structurally matched dynamics track with 8 paired seeds and equal-budget hyperparameter sweeps. Every trained model failed the certificate and used the identical fixed-point fallback, producing exactly the baseline MSE; the mechanism prediction was not confirmed.", "metrics": { "baseline": "Test MSE mean 0.0042600604, std 0.0017312099, best lr=0.006 and iters=24", "idea": "Test MSE mean 0.0042600604, std 0.0017312099, best lr=0.006 and iters=24; paired delta 0.0, permutation p=1.0; observed q mean 3.8198, range 3.5949-4.2700, certified cases 0/8" }, "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 and a small implicit recurrent architecture were tested. No higher-dimensional DEQ, parameter/input interval subdivision, wall-clock inference comparison, or trained contractive regime with successful certificates was observed. The local bounds are conservative analytic bounds rather than a full directed-rounding interval library.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "custom_equilibrium_rnn", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "iters": 24 }, "sweep": [ { "cfg": { "lr": 0.001, "iters": 8 }, "mean": 0.04040969908237457 }, { "cfg": { "lr": 0.003, "iters": 16 }, "mean": 0.008657814585603774 }, { "cfg": { "lr": 0.006, "iters": 24 }, "mean": 0.004971429007127881 } ], "full": { "mean": 0.004260060406522825, "std": 0.0017312099173934859, "per_seed": [ 0.00337711232714355, 0.007119504269212484, 0.007134149316698313, 0.002254950115457177, 0.003825985360890627, 0.003774178447201848, 0.0037776767276227474, 0.0028169266879558563 ], "n": 8 } }, "idea": { "mean": 0.004260060406522825, "std": 0.0017312099173934859, "per_seed": [ 0.00337711232714355, 0.007119504269212484, 0.007134149316698313, 0.002254950115457177, 0.003825985360890627, 0.003774178447201848, 0.0037776767276227474, 0.0028169266879558563 ], "n": 8, "best_cfg": { "lr": 0.006, "iters": 24 }, "sweep": [ { "cfg": { "lr": 0.001, "iters": 8 }, "mean": 0.044312840327620506 }, { "cfg": { "lr": 0.003, "iters": 16 }, "mean": 0.008995423908345401 }, { "cfg": { "lr": 0.006, "iters": 24 }, "mean": 0.004260060406522825 } ] }, "comparison": { "delta_mean": 0.0, "idea_wins": 0, "n_pairs": 8, "per_seed_diffs": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "p_value": 1.0, "mde": 0.0, "mde_rel_pct": 0.0, "verdict": "no measurable effect", "system_worked": false }, "mechanism_signature": { "prediction": "trained-model local q below 0.8 predicts certification; q near or above 1 predicts fallback", "n_models": 8, "predicted_low_q_count": 0, "observed_certified_low_q_count": 0, "mean_q": 3.819756507873535, "min_q": 3.59490966796875, "max_q": 4.270003318786621, "q_values": [ 3.786127805709839, 3.7302780151367188, 3.6078004837036133, 3.59490966796875, 3.9719083309173584, 4.270003318786621, 3.7772746086120605, 3.8197498321533203 ], "certified_flags": [ 0, 0, 0, 0, 0, 0, 0, 0 ], "confirmed": false }, "architecture_note": "Both arms train the same implicit tanh recurrent layer; baseline always iterates, idea certifies then falls back.", "runtime_sec": 50.54426717758179 }, "system_verdict": "failed", "practical_verdict": "no_effect", "mechanism_ok": 0, "system_judged": true }