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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Implemented residual-gated polynomial lift features around a matched GRU on the structurally appropriate controlled-pendulum dynamics track. The trained-model cubic proxy scaled by 3.375 under 1.5x input amplitude, matching 1.5^3, but the gate was active on 99.5% of test examples. The idea was significantly worse: delta_mean=+0.0006127432 MSE, permutation p=0.0324, so worked=false.",
  "metrics": {
    "baseline": "Fixed degree-2 lifted GRU, best lr=0.006: mean test MSE 0.001581461, std 0.000439250, 8 paired seeds.",
    "idea": "Residual-gated cubic lifted GRU, best lr=0.006 threshold=0.4: mean test MSE 0.002194204, std 0.000755785, 2/8 paired wins; delta_mean=+0.0006127432, p=0.0324.",
    "mechanism_signature": {
      "confirmed": true,
      "prediction": "cubic residual proxy scales as amplitude^3",
      "observed_mean_proxy_ratio_at_1p5": 3.375,
      "predicted_ratio_at_1p5": 3.375,
      "mean_rho": 0.8967330754,
      "mean_active_fraction": 0.9953125,
      "track_match": "controlled-pendulum dynamics; standard test MSE on independently trained systems"
    }
  },
  "bench_report": {
    "bench_version": 1,
    "track": "dynamics",
    "model": "rnn_small",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.006,
        "degree": 2
      },
      "full": {
        "mean": 0.0015814606886124238,
        "std": 0.0004392499281827401,
        "per_seed": [
          0.0010742004960775375,
          0.0021969457156956196,
          0.0015373978530988097,
          0.002299107611179352,
          0.001275557209737599,
          0.0012933816760778427,
          0.0011743184877559543,
          0.0018007764592766762
        ],
        "n": 8
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001,
            "degree": 2
          },
          "mean": 0.003609913313994184
        },
        {
          "cfg": {
            "lr": 0.003,
            "degree": 2
          },
          "mean": 0.004211013438180089
        },
        {
          "cfg": {
            "lr": 0.006,
            "degree": 2
          },
          "mean": 0.0017769129190128297
        },
        {
          "cfg": {
            "lr": 0.001,
            "degree": 3
          },
          "mean": 0.02138834842480719
        },
        {
          "cfg": {
            "lr": 0.003,
            "degree": 3
          },
          "mean": 0.005644599266815931
        },
        {
          "cfg": {
            "lr": 0.006,
            "degree": 3
          },
          "mean": 0.0025558681518305093
        }
      ]
    },
    "idea": {
      "best_cfg": {
        "lr": 0.006,
        "threshold": 0.4
      },
      "mean": 0.0021942039020359516,
      "std": 0.0007557853669569652,
      "per_seed": [
        0.0022852420806884766,
        0.002649907488375902,
        0.001380756264552474,
        0.0038659078531718253,
        0.0015587329860773087,
        0.002379080746695299,
        0.0016722837554104628,
        0.0011597189774364234
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 0.0006127432134235278,
      "idea_wins": 2,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.001211041584610939,
        0.0004529617726802826,
        -0.0001566415885463357,
        0.0015668005216866732,
        0.00028317479882389307,
        0.0010856990702450275,
        0.0004979649093002081,
        -3.905536141246557e-05
      ],
      "p_value": 0.0324,
      "mde": 0.0005150508833037192,
      "mde_rel_pct": 32.5680484511838,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "confirmed": true,
      "prediction": "cubic residual proxy scales as amplitude^3",
      "observed_mean_proxy_ratio_at_1p5": 3.375,
      "mean_rho": 0.8967330753803253,
      "mean_active_fraction": 0.9953124970197678
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_residual_gated.py",
  "files": [
    "bench_residual_gated.py",
    "bench_report.json",
    "run.log"
  ],
  "limitations": "The intervention used a normalized cubic feature proxy rather than finite-difference observed derivatives or a learned closure; no noisy-observation robustness, wall-clock/FLOP accounting, or longer rollout benchmark was tested. The baseline and idea used 12 epochs on the standard 400/160 dynamics split.",
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}