Intrinsic-Rank Filter Memory for Actor-Critic / report_bench_2026-09-01T125339.md

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP

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

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 9, "verdict": "Built the intrinsic-rank filter-memory intervention on the structurally matched dynamics track using a canonical rnn_small-style GRU, causal filtering, and train-split SVD bottlenecking. The mechanism signature was confirmed: ambient width 9 reduced to observed rank 6, matching the predicted MIMO intrinsic rank. However, the eight-seed paired result was not significant (delta_mean=-3.8747e-05, p=0.8938), so the idea does not meet the benchmark win criterion.", "metrics": { "baseline": "Test MSE 0.0011299664 ± 0.0003363602 over 8 seeds; best lr=0.01.", "idea": "Test MSE 0.0010912194 ± 0.0004496559 over 8 seeds; best lr=0.01, tau=0.01; paired delta=-3.8746963e-05, p=0.8938, 5/8 wins." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py", "files": [ "stage2_bench.py", "bench_report.json", "bench_stdout.txt" ], "limitations": "This tests supervised pendulum prediction rather than actor-critic policy learning or Bellman regression. The lift deliberately included duplicated channels to create deterministic null directions; online SVD updates, Procrustes alignment, compute/memory speed, and true closed-loop actor-critic performance were not tested.", "bench_report": { "bench_version": 1, "track": "dynamics", "model": "rnn_small", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.01, "tau": 0.01 }, "full": { "mean": 0.0011299664110993035, "std": 0.00033636016341449245, "per_seed": [ 0.0008199093863368034, 0.0008286746451631188, 0.0016839810414239764, 0.0009439851273782551, 0.001039800001308322, 0.0017109999898821115, 0.0009939224692061543, 0.0010184586280956864 ], "n": 8 } }, "idea": { "best_cfg": { "lr": 0.01, "tau": 0.01 }, "mean": 0.0010912194484262727, "std": 0.00044965589408538725, "per_seed": [ 0.0005372872692532837, 0.0008794514578767121, 0.001270528300665319, 0.0019876640290021896, 0.0010541900992393494, 0.0015042235609143972, 0.0008550031343474984, 0.0006414077361114323 ], "n": 8 }, "comparison": { "delta_mean": -3.8746962673030794e-05, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ -0.0002826221170835197, 5.0776812713593245e-05, -0.00041345274075865746, 0.0010436789016239345, 1.439009793102741e-05, -0.0002067764289677143, -0.00013891933485865593, -0.0003770508919842541 ], "p_value": 0.8938, "mde": 0.0003915640556028904, "mde_rel_pct": 34.65271637782152, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "prediction": "causal lifted ambient width 9 has deterministic duplicate directions and SVD rank 6", "predicted_intrinsic_rank": 6, "observed_rank_tau_1e-3": 6, "ambient_coordinate_width": 9, "trained_model_behavior": { "baseline": { "metric": 0.00200747512280941, "prediction_rms_under_input_perturbation": 0.0009409512858837843 }, "idea": { "metric": 0.0012641767971217632, "prediction_rms_under_input_perturbation": 0.0011408096179366112 } }, "confirmed": true } }, "system_verdict": "partial", "practical_verdict": "inconclusive", "mechanism_ok": 1, "system_judged": true }