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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "The Contact-Splitting Momentum Optimizer was implemented and evaluated on the registered optimizer-appropriate tabular track with the same mlp_tiny architecture and equal gradient budgets. It was significantly worse than tuned momentum: mean test MSE 82.4742 versus 6.1027, paired delta +76.3715, permutation p=0.0081. The trained-model certificate-rate signature was not confirmed, with observed rate 0.2076 versus predicted gamma 0.1.",
  "metrics": {
    "baseline": "mean test MSE 6.102685570716858, std 1.0114630114444267, best lr=0.012 beta=0.85",
    "idea": "mean test MSE 82.47423553466797, std 62.85156905560213, best lr=0.003 gamma=0.1",
    "paired_delta": "76.37154996395111",
    "permutation_p_value": "0.0081"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_contact.py",
  "files": [
    "bench_contact.py",
    "bench_report.json"
  ],
  "limitations": "Only the registered tabular track was tested. The MVP used identity mass M and fixed gamma=0.1; learned diagonal preconditioning and wall-clock timing were not tested.",
  "bench_report": {
    "bench_version": 1,
    "track": "tabular",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.012,
        "beta": 0.85
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.003,
            "beta": 0.85
          },
          "mean": 7.4412044286727905
        },
        {
          "cfg": {
            "lr": 0.003,
            "beta": 0.95
          },
          "mean": 7.356154918670654
        },
        {
          "cfg": {
            "lr": 0.006,
            "beta": 0.85
          },
          "mean": 7.447907209396362
        },
        {
          "cfg": {
            "lr": 0.006,
            "beta": 0.95
          },
          "mean": 6.9799922704696655
        },
        {
          "cfg": {
            "lr": 0.012,
            "beta": 0.85
          },
          "mean": 6.368864178657532
        },
        {
          "cfg": {
            "lr": 0.012,
            "beta": 0.95
          },
          "mean": 11.507428050041199
        }
      ],
      "full": {
        "mean": 6.102685570716858,
        "std": 1.0114630114444267,
        "per_seed": [
          5.218059539794922,
          7.1943206787109375,
          6.8049211502075195,
          6.258155345916748,
          4.909553527832031,
          4.723381042480469,
          7.64085054397583,
          6.072242736816406
        ],
        "n": 8,
        "details": [
          {
            "metric": 5.218059539794922,
            "grad_evals": 252
          },
          {
            "metric": 7.1943206787109375,
            "grad_evals": 252
          },
          {
            "metric": 6.8049211502075195,
            "grad_evals": 252
          },
          {
            "metric": 6.258155345916748,
            "grad_evals": 252
          },
          {
            "metric": 4.909553527832031,
            "grad_evals": 252
          },
          {
            "metric": 4.723381042480469,
            "grad_evals": 252
          },
          {
            "metric": 7.64085054397583,
            "grad_evals": 252
          },
          {
            "metric": 6.072242736816406,
            "grad_evals": 252
          }
        ]
      }
    },
    "idea": {
      "mean": 82.47423553466797,
      "std": 62.85156905560213,
      "per_seed": [
        85.1873779296875,
        61.3437385559082,
        54.8747444152832,
        16.73917007446289,
        198.2187042236328,
        37.13703918457031,
        172.99974060058594,
        33.29336929321289
      ],
      "n": 8,
      "details": [
        {
          "metric": 85.1873779296875,
          "grad_evals": 252,
          "cert_rate_mean": 0.3926487951538228,
          "cert_rate_n": 125
        },
        {
          "metric": 61.3437385559082,
          "grad_evals": 252,
          "cert_rate_mean": 0.2918240217992533,
          "cert_rate_n": 125
        },
        {
          "metric": 54.8747444152832,
          "grad_evals": 252,
          "cert_rate_mean": -0.11212525637936967,
          "cert_rate_n": 125
        },
        {
          "metric": 16.73917007446289,
          "grad_evals": 252,
          "cert_rate_mean": 0.1972534224451282,
          "cert_rate_n": 125
        },
        {
          "metric": 198.2187042236328,
          "grad_evals": 252,
          "cert_rate_mean": 0.15344815565986625,
          "cert_rate_n": 125
        },
        {
          "metric": 37.13703918457031,
          "grad_evals": 252,
          "cert_rate_mean": 0.0256587405843517,
          "cert_rate_n": 125
        },
        {
          "metric": 172.99974060058594,
          "grad_evals": 252,
          "cert_rate_mean": 0.4365923490884733,
          "cert_rate_n": 125
        },
        {
          "metric": 33.29336929321289,
          "grad_evals": 252,
          "cert_rate_mean": 0.2754857404526831,
          "cert_rate_n": 125
        }
      ]
    },
    "comparison": {
      "delta_mean": 76.37154996395111,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        79.96931838989258,
        54.149417877197266,
        48.069823265075684,
        10.481014728546143,
        193.30915069580078,
        32.413658142089844,
        165.3588900566101,
        27.221126556518556
      ],
      "p_value": 0.0081,
      "mde": 56.17144123095511,
      "mde_rel_pct": 920.4380691099063,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "idea_sweep": [
        {
          "cfg": {
            "lr": 0.003,
            "gamma": 0.1
          },
          "mean": 54.53625774383545
        },
        {
          "cfg": {
            "lr": 0.006,
            "gamma": 0.1
          },
          "mean": 76.64579582214355
        },
        {
          "cfg": {
            "lr": 0.012,
            "gamma": 0.1
          },
          "mean": 63.562326431274414
        }
      ],
      "mechanism_signature": {
        "prediction": "certificate conformal rate ~= gamma",
        "predicted": 0.1,
        "observed": 0.20759824610052613,
        "absolute_error": 0.10759824610052612,
        "n_models": 8,
        "confirmed": false
      },
      "budget": {
        "epochs": 18,
        "batch": 64,
        "baseline_gradient_evals": 252,
        "idea_gradient_evals": 252
      }
    }
  },
  "system_verdict": "failed",
  "practical_verdict": "harms",
  "mechanism_ok": 0,
  "system_judged": true
}