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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "The registered tabular Friedman regression benchmark used mlp_tiny, a tuned Adam baseline sweep, shared learning-rate parity, and 8 paired seeds. The negative-shift idea had lower mean MSE but was not a significant win (delta_mean=-0.47317224740982056, p_value=0.1835); the selected configuration used nu=0 and all nonzero shifts were worse. The trained-model mechanism signature was not confirmed.",
  "metrics": {
    "baseline": "Adam best cfg lr=0.01, weight_decay=0.0001; mean test MSE=8.206986248493195.",
    "idea": "Negative-shift Adam best cfg lr=0.01, weight_decay=0.0001, nu_mult=0.0; mean test MSE=7.733814001083374; paired delta=-0.47317224740982056; p=0.1835."
  },
  "bench_report": {
    "bench_version": 1,
    "track": "tabular",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01,
        "weight_decay": 0.0001
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001,
            "weight_decay": 0.0
          },
          "mean": 25.60164165496826
        },
        {
          "cfg": {
            "lr": 0.001,
            "weight_decay": 0.0001
          },
          "mean": 25.508389472961426
        },
        {
          "cfg": {
            "lr": 0.003,
            "weight_decay": 0.0
          },
          "mean": 13.860755920410156
        },
        {
          "cfg": {
            "lr": 0.003,
            "weight_decay": 0.0001
          },
          "mean": 13.839483976364136
        },
        {
          "cfg": {
            "lr": 0.01,
            "weight_decay": 0.0
          },
          "mean": 8.352817177772522
        },
        {
          "cfg": {
            "lr": 0.01,
            "weight_decay": 0.0001
          },
          "mean": 8.268931150436401
        }
      ],
      "full": {
        "mean": 8.206986248493195,
        "std": 0.7130848431694788,
        "per_seed": [
          7.954751968383789,
          9.628533363342285,
          7.705244541168213,
          7.787194728851318,
          7.160106182098389,
          8.077467918395996,
          8.747560501098633,
          8.595030784606934
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 7.733814001083374,
      "std": 0.6299701216517714,
      "per_seed": [
        6.852144241333008,
        7.3872199058532715,
        7.651295185089111,
        8.405105590820312,
        7.134853839874268,
        7.516578197479248,
        8.040628433227539,
        8.882686614990234
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": -0.47317224740982056,
      "idea_wins": 6,
      "n_pairs": 8,
      "per_seed_diffs": [
        -1.1026077270507812,
        -2.2413134574890137,
        -0.05394935607910156,
        0.6179108619689941,
        -0.025252342224121094,
        -0.560889720916748,
        -0.7069320678710938,
        0.2876558303833008
      ],
      "p_value": 0.1835,
      "mde": 0.756546047206517,
      "mde_rel_pct": 9.218317471232746,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "predicted": {
        "one_step_displacement_factor": 1.089914695739746,
        "nu": 8.991469573974609,
        "curvature_scale": 59.94313049316406
      },
      "observed": {
        "trained_positive_shift_disp": 50.0323600769043,
        "trained_zero_shift_disp": 7.432165145874023,
        "whole_run_ratio": 6.731868721281016
      },
      "confirmed": false
    },
    "idea_sweep": {
      "best_cfg": {
        "lr": 0.01,
        "weight_decay": 0.0001,
        "nu_mult": 0.0
      },
      "shared_lr_union": [
        0.001,
        0.003,
        0.01
      ],
      "nu_multipliers": [
        0.0,
        0.05,
        0.15
      ]
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_negative_shift_bench.py",
  "files": [
    "stage2_negative_shift_bench.py",
    "bench_report.json"
  ],
  "limitations": "Only the registered tabular Friedman regression track with mlp_tiny was tested. The benchmark used minibatch Adam with a displacement-centered negative term rather than exact full-batch Euler dynamics; the harness provided no validation split; and the mechanism signature compared a one-step predicted factor with a whole-run observed displacement ratio.",
  "system_verdict": "failed",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 0,
  "system_judged": true
}