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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "The registered potential_transport track was verified in bench.all_track_names(), and the full 8-seed paired protocol with tuned baseline sweep was rerun. Exact OT pairing confirmed the mechanism through a 93.5% reduction in trained-model trajectory curvature, but test MSE was significantly worse (paired delta +0.1878167, permutation p=0.0081), so this is not a benchmark win.",
  "metrics": {
    "baseline": "Mean test MSE 1.1834656298160553; best lr=0.003, epochs=24.",
    "idea": "Mean test MSE 1.3712822943925858; best lr=0.003, epochs=24; paired delta +0.18781666457653046; p=0.0081; 0/8 wins.",
    "mechanism_signature": "Baseline curvature 0.0028906640696472357; idea curvature 0.00018671443935350647; relative reduction 0.9354077696837695; confirmed=true."
  },
  "bench_report": {
    "bench_version": 1,
    "track": "potential_transport",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.003,
        "epochs": 24
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001,
            "epochs": 24
          },
          "mean": 1.2874779403209686
        },
        {
          "cfg": {
            "lr": 0.003,
            "epochs": 24
          },
          "mean": 1.2407947480678558
        },
        {
          "cfg": {
            "lr": 0.009,
            "epochs": 24
          },
          "mean": 1.2557097375392914
        }
      ],
      "full": {
        "mean": 1.1834656298160553,
        "std": 0.11952528701329043,
        "per_seed": [
          1.110816478729248,
          1.1996243000030518,
          1.3317530155181885,
          1.320985198020935,
          1.2276966571807861,
          1.0896412134170532,
          0.9505289793014526,
          1.236679196357727
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 1.3712822943925858,
      "std": 0.1065033848960024,
      "per_seed": [
        1.2836636304855347,
        1.4103634357452393,
        1.4827837944030762,
        1.4776341915130615,
        1.4404963254928589,
        1.2114852666854858,
        1.2215279340744019,
        1.4423037767410278
      ],
      "n": 8,
      "best_cfg": {
        "lr": 0.003,
        "epochs": 24
      }
    },
    "comparison": {
      "delta_mean": 0.18781666457653046,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.17284715175628662,
        0.2107391357421875,
        0.1510307788848877,
        0.15664899349212646,
        0.21279966831207275,
        0.12184405326843262,
        0.2709989547729492,
        0.20562458038330078
      ],
      "p_value": 0.0081,
      "mde": 0.03907506574222071,
      "mde_rel_pct": 3.3017490967011947,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001,
          "epochs": 24
        },
        "mean": 1.4397288411855698,
        "std": 0.10629865170104423
      },
      {
        "cfg": {
          "lr": 0.003,
          "epochs": 24
        },
        "mean": 1.3712822943925858,
        "std": 0.1065033848960024
      },
      {
        "cfg": {
          "lr": 0.009,
          "epochs": 24
        },
        "mean": 1.3912305235862732,
        "std": 0.10868632423878194
      }
    ],
    "track_selection": "Exact continuous transport / flow matching structure; source and target endpoints are reconstructed from each bridge sample.",
    "custom_track": {
      "name": "potential_transport",
      "file": "bench/custom_tracks/potential_transport.py",
      "domain": "continuous transport / flow matching"
    },
    "mechanism_signature": {
      "prediction": "OT-reflow training should produce straighter learned trajectories, measured by lower eight-step velocity second-difference.",
      "predicted_curvature_idea_lt_baseline": true,
      "observed_baseline_curvature": 0.0028906640696472357,
      "observed_idea_curvature": 0.00018671443935350647,
      "relative_reduction": 0.9354077696837695,
      "confirmed": true
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_experiment.py",
  "files": [
    "bench_experiment.py",
    "bench_report.json"
  ],
  "limitations": "Only the registered potential_transport custom track was tested; CIFAR-10, larger flow models, Sinkhorn approximations, multiple outer reflow rounds, Euler sampling-quality curves, FID, and wall-clock/FLOP measurements were not tested. The track permits exact endpoint reconstruction, so this tests minibatch OT re-pairing but not a fully independent learned ODE endpoint-regeneration outer loop.",
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}