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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Built and evaluated Residual-Relaxed Trajectory Sampling on the structurally matched dynamics track with shared rnn_small systems, paired seeds, and equal-budget learning-rate sweeps. The trained-model mechanism signature was confirmed because high-residual cases had flatter candidate weights (ESS 28.1731 versus 25.5943). The task-metric improvement was not significant (delta_mean=-2.2066e-05, p=0.44995), so this is not a benchmark win.",
  "metrics": {
    "baseline": "Test MSE mean 0.0005640739 across 8 seeds; best lr=0.006.",
    "idea": "Test MSE mean 0.0005420081 across 8 seeds; best lr=0.006, alpha=2.0, beta=0.9, sigma=0.05; paired delta=-2.2066e-05, p=0.44995, 4/8 wins.",
    "mechanism_signature": "Residual mean 0.0195247, p90 0.0324388; ESS low-residual 25.5943, high-residual 28.1731, ratio 1.10076; confirmed=true."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_stage2.py",
  "files": [
    "bench_stage2.py",
    "bench_report.json"
  ],
  "limitations": "Only the built-in dynamics track was tested, using 8 epochs, 1200 training samples, 400 test samples, three shared learning-rate values, and three candidate perturbations. This implementation is a supervised training-loop surrogate for residual-adaptive candidate sampling rather than a deployed learned-model MPPI controller; ensemble disagreement, longer horizons, and other tracks 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": 0.006
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.0015
          },
          "mean": 0.0011108634
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.000810568
        },
        {
          "cfg": {
            "lr": 0.006
          },
          "mean": 0.0006289426
        }
      ],
      "full": {
        "mean": 0.0005640739,
        "std": 0.0001367984,
        "per_seed": [
          0.0004853792,
          0.0007160367,
          0.00049965,
          0.0008147045,
          0.0004540994,
          0.0005191828,
          0.0003816201,
          0.0006419186
        ],
        "n": 8
      }
    },
    "idea": {
      "best_cfg": {
        "lr": 0.006,
        "alpha": 2.0,
        "beta": 0.9,
        "sigma": 0.05
      },
      "mean": 0.0005420081,
      "std": 0.0001396002,
      "per_seed": [
        0.0005091279,
        0.000558412,
        0.0004696174,
        0.0008545062,
        0.0004419976,
        0.0003986097,
        0.0004496737,
        0.0006541206
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": -2.20658e-05,
      "idea_wins": 4,
      "n_pairs": 8,
      "p_value": 0.44995,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "Larger trained-model observed residual induces larger temperature and flatter candidate weights.",
      "observed_residual_mean": 0.0195247,
      "observed_residual_p90": 0.0324388,
      "observed_ESS_low_residual": 25.5943,
      "observed_ESS_high_residual": 28.1731,
      "ESS_ratio_high_over_low": 1.10076,
      "confirmed": true
    },
    "protocol_notes": {
      "epochs": 8,
      "n_train": 1200,
      "n_test": 400,
      "batch": 128,
      "track_choice": "dynamics matches stability/control and multi-step pendulum rollouts",
      "lr_union": [
        0.0015,
        0.003,
        0.006
      ]
    }
  },
  "system_verdict": "partial",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 1,
  "system_judged": true
}