OT Primitive Universal Flow / report_bench_2026-09-02T162717.md

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP

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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Implemented the OT-inspired intervention on the registered dynamics track with the shared rnn_small architecture, adding monotonicity and displacement penalties to MSE. The idea was slightly worse than baseline and not significant: paired delta_mean was +2.777924237307161e-7 and permutation p=0.39955. The trained-model positive-Jacobian signature was confirmed, but it did not yield a task-metric win.", "metrics": { "baseline": "8-seed mean test MSE 0.0005214317206991836, std 0.000071050613588211; best lr=0.006", "idea": "8-seed mean test MSE 0.0005217095131229144, std 0.000071353424012946; best lr=0.006; 2/8 paired wins; delta_mean +2.777924237307161e-7; p=0.39955" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_ot_dynamics.py", "files": [ "bench_ot_dynamics.py", "bench_report.json", "bench_run.log" ], "limitations": "Only the registered built-in dynamics track was tested. No manifold exponential map, full density-to-density OT likelihood, RealNVP, neural ODE, image/sequence tracks, or larger models were evaluated. The benchmark proxy constrains the terminal-angle Jacobian rather than learning a complete density transport map.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0 }, "mean": 0.001414267288055271 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0008843674877425656 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 0.0005579828139161691 } ], "full": { "mean": 0.0005214317206991836, "std": 7.1050613588211e-05, "per_seed": [ 0.0005374288302846253, 0.00047665066085755825, 0.0005716764717362821, 0.0006461752927862108, 0.0005541553255170584, 0.0004454980371519923, 0.00040709308814257383, 0.0005327760591171682 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.006, "weight_decay": 0.0 }, "sweep": [ { "cfg": { "lr": 0.0015, "weight_decay": 0.0 }, "mean": 0.0013487355754477903, "std": 0.0004249648894380296, "n": 8 }, { "cfg": { "lr": 0.003, "weight_decay": 0.0 }, "mean": 0.0008453956179437228, "std": 0.00020597588608945252, "n": 8 }, { "cfg": { "lr": 0.006, "weight_decay": 0.0 }, "mean": 0.0005217095131229144, "std": 7.1353424012946e-05, "n": 8 } ], "per_seed": [ 0.0005381343653425574, 0.0004766805504914373, 0.000572634453419596, 0.0006471937522292137, 0.0005527037428691983, 0.0004458390176296234, 0.00040668679866939783, 0.0005338034243322909 ], "mean": 0.0005217095131229144, "std": 7.1353424012946e-05, "n": 8 }, "comparison": { "delta_mean": 2.777924237307161e-07, "idea_wins": 2, "n_pairs": 8, "per_seed_diffs": [ 7.055350579321384e-07, 2.9889633879065514e-08, 9.579816833138466e-07, 1.0184594430029392e-06, -1.4515826478600502e-06, 3.4098047763109207e-07, -4.062894731760025e-07, 1.0273652151226997e-06 ], "p_value": 0.39955, "verdict": "no measurable effect", "system_worked": false }, "mechanism_signature": { "prediction": "OT monotonicity predicts positive terminal-angle Jacobian", "idea": { "predicted_monotone_fraction": 1.0, "observed_monotone_fraction": 1.0, "observed_mean_terminal_angle_jacobian": 0.24786332249641418, "confirmed": true }, "baseline": { "observed_monotone_fraction": 1.0, "observed_mean_terminal_angle_jacobian": 0.2478588968515396 }, "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "no_effect", "mechanism_ok": 1, "system_judged": true }