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

(универсальный)

Ответ агента:

{
  "worked": true,
  "confidence": 8,
  "verdict": "The inverse-gain structured GRU significantly outperformed the matched direct-action GRU on the registered inverse_gain_control dynamics track. Test MSE improved from 3.6709986031 to 1.5069266558 on all 8 paired seeds, with permutation p=0.0081; the trained-model mechanism signature confirmed the predicted action-error factorization.",
  "metrics": {
    "baseline": "Direct-action GRU, tuned lr=0.003: mean test MSE 3.670998603105545, std 0.63283622936479.",
    "idea": "Inverse-gain structured GRU, tuned lr=0.003: mean test MSE 1.5069266557693481, std 0.23555139171685704.",
    "paired_delta": "-2.164071947336197; 8/8 idea wins; permutation p=0.0081."
  },
  "bench_report": {
    "bench_version": 1,
    "track": "inverse_gain_control",
    "model": "rnn_small",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.003
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 5.920182824134827
        },
        {
          "cfg": {
            "lr": 0.002
          },
          "mean": 4.574557304382324
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 3.198916256427765
        }
      ],
      "full": {
        "mean": 3.670998603105545,
        "std": 0.63283622936479,
        "per_seed": [
          3.7461042404174805,
          3.104163408279419,
          2.9681482315063477,
          2.9772491455078125,
          4.229879379272461,
          4.706507682800293,
          4.304690361022949,
          3.3312463760375977
        ],
        "n": 8
      }
    },
    "idea": {
      "best_cfg": {
        "lr": 0.003
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 2.0682411789894104
        },
        {
          "cfg": {
            "lr": 0.002
          },
          "mean": 1.5857862532138824
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 1.318423479795456
        }
      ],
      "mean": 1.5069266557693481,
      "std": 0.23555139171685704,
      "per_seed": [
        1.5180333852767944,
        1.335729956626892,
        1.2053886651992798,
        1.2145419120788574,
        1.7869166135787964,
        1.6817357540130615,
        1.8618539571762085,
        1.451213002204895
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": -2.164071947336197,
      "idea_wins": 8,
      "n_pairs": 8,
      "per_seed_diffs": [
        -2.228070855140686,
        -1.7684334516525269,
        -1.7627595663070679,
        -1.762707233428955,
        -2.4429627656936646,
        -3.0247719287872314,
        -2.4428364038467407,
        -1.8800333738327026
      ],
      "p_value": 0.0081,
      "mde": 0.3818346490175283,
      "mde_rel_pct": 10.40138366422991,
      "verdict": "idea better (significant)",
      "system_worked": true
    },
    "custom_track": {
      "name": "inverse_gain_control",
      "file": "custom_inverse_gain_track.py",
      "domain": "dynamics"
    },
    "mechanism_signature": {
      "n_test": 400,
      "observed_action_error_mean_abs": 0.7880314588546753,
      "predicted_factor_error_mean_abs": 0.7880314588546753,
      "slope_observed_on_predicted": 1.0,
      "correlation": 0.9999999999999957,
      "relative_residual": 9.462353034450643e-08,
      "inverse_gain_mae": 0.8039304614067078,
      "prediction": "action_error=(qhat-q_observed)*z",
      "confirmed": true
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_run.py",
  "files": [
    "custom_inverse_gain_track.py",
    "bench_run.py",
    "bench_report.json",
    "/home/maxwelhelp/all/math2nn/bench/custom_tracks/inverse_gain_control.py"
  ],
  "limitations": "The registered custom track is a small supervised sampled-control benchmark and does not test long-horizon closed-loop stability, sensor-noise robustness, actuator saturation, or hardware transfer. The mechanism signature uses privileged expert targets to infer observed inverse gain and therefore validates the neural-scale factorization rather than deployment-time gain estimation.",
  "system_verdict": "worked",
  "practical_verdict": "helps",
  "mechanism_ok": 1,
  "system_judged": true
}