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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Built matched two-component MDN systems on a custom structurally appropriate bimodal regression track, comparing Monte Carlo entropy regularization with deterministic Gauss–Hermite entropy regularization. GH reduced mean test NLL from 0.760407 to 0.730448, but the paired permutation test gave p=0.7841, so there is no significant benchmark win. The trained-model mechanism signature was confirmed: GH repeated-estimator standard deviation was zero while 32-sample MC entropy had nonzero standard deviation.",
  "metrics": {
    "baseline": "MC entropy regularization, tuned lr=0.006: mean test NLL 0.760407 over 8 seeds",
    "idea": "GH K=7 entropy regularization: mean test NLL 0.730448 over 8 seeds; paired delta -0.029959, p=0.7841, 5/8 wins"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 mdn_bench.py",
  "files": [
    "mdn_bench.py",
    "mdn_track.py",
    "bench_report.json"
  ],
  "limitations": "Only one custom 2-D diagonal-covariance MDN task was tested. Full-covariance heads, dimensions above two, broader entropy-coefficient sweeps, wall-clock speed, calibration, and external datasets were not tested. The mechanism signature used two trained-model seeds; the primary comparison used the required eight paired seeds.",
  "bench_report": {
    "bench_version": 1,
    "track": "custom_track:conditional_mixture_entropy_regression",
    "model": "mdn_shared",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "grid": [
        {
          "lr": 0.001
        },
        {
          "lr": 0.003
        },
        {
          "lr": 0.006
        }
      ],
      "best_config": {
        "lr": 0.006
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 1.141812
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.7959
        },
        {
          "cfg": {
            "lr": 0.006
          },
          "mean": 0.769664
        }
      ],
      "full": {
        "mean": 0.760407,
        "std": 0.246538,
        "per_seed": [
          0.615586,
          0.611552,
          0.706548,
          1.14497,
          0.581347,
          0.605724,
          0.598669,
          1.218861
        ],
        "n": 8
      }
    },
    "idea": {
      "method": "deterministic componentwise Gauss-Hermite entropy, K=7",
      "mean": 0.730448,
      "std": 0.188653,
      "per_seed": [
        0.708852,
        0.608228,
        0.66842,
        0.792318,
        0.616417,
        0.680534,
        0.56918,
        1.199638
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": -0.029959,
      "idea_wins": 5,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.093267,
        -0.003323,
        -0.038128,
        -0.352652,
        0.03507,
        0.074809,
        -0.029489,
        -0.019223
      ],
      "p_value": 0.7841,
      "mde": 0.116296,
      "mde_rel_pct": 15.293954,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "Fixed Gauss-Hermite entropy has zero estimator repeat variance while finite-sample Monte Carlo entropy has nonzero variance.",
      "observed": [
        {
          "seed": 0,
          "gh_entropy": 0.549249,
          "gh_repeat_std": 0.0,
          "mc_repeat_std": 0.006768
        },
        {
          "seed": 1,
          "gh_entropy": 0.558372,
          "gh_repeat_std": 0.0,
          "mc_repeat_std": 0.01105
        }
      ],
      "confirmed": true
    },
    "custom_track": {
      "name": "conditional_mixture_entropy_regression",
      "file": "mdn_track.py",
      "domain": "probabilistic-regression"
    }
  },
  "system_verdict": "partial",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 1,
  "system_judged": true
}