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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Built and ran a custom SO(3)-structured regression track because no built-in track contains orientation-field differential structure. The exact-generator system was slightly worse than the matched finite-difference baseline (MSE 0.014595 vs 0.014346), with paired delta +0.000249 and permutation p=0.57535, so no transfer win was observed.",
  "metrics": {
    "baseline": "8-seed tuned MSE 0.0143462 ± 0.0012082; best lr=0.003, h=0.2, epochs=12.",
    "idea": "8-seed exact-generator MSE 0.0145949 ± 0.0017917; paired delta +0.0002487, 3/8 wins, permutation p=0.57535."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 run_bench.py",
  "files": [
    "so3_custom_track.py",
    "run_bench.py",
    "bench_report.json"
  ],
  "limitations": "This is a compact coefficient-regression proxy, not a full SO(3) FFT/IFFT pose or molecular model. Wall-clock speed, trained-model equivariance, vector divergence heads, sparse GPU kernels, and real Wigner-D symmetry were not tested.",
  "bench_report": {
    "bench_version": 1,
    "track": "so3_spectral_regression",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.003,
        "h": 0.2,
        "epochs": 12
      },
      "full": {
        "mean": 0.01434623496606946,
        "std": 0.0012081506817864598,
        "per_seed": [
          0.013110446743667126,
          0.012636005878448486,
          0.016413535922765732,
          0.014096695929765701,
          0.013522281311452389,
          0.014381379820406437,
          0.01573561690747738,
          0.01487391721457243
        ],
        "n": 8
      }
    },
    "idea": {
      "best_cfg": {
        "lr": 0.003,
        "h": 0.2,
        "epochs": 12
      },
      "full": {
        "mean": 0.014594940701499581,
        "std": 0.0017917110895247232,
        "per_seed": [
          0.013325662352144718,
          0.01207823771983385,
          0.016623985022306442,
          0.01392221637070179,
          0.014122492633759975,
          0.013632146641612053,
          0.01803544908761978,
          0.01501933578401804
        ],
        "n": 8
      }
    },
    "comparison": {
      "delta_mean": 0.0002487057354301214,
      "idea_wins": 3,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.00021521560847759247,
        -0.0005577681586146355,
        0.00021044909954071045,
        -0.00017447955906391144,
        0.0006002113223075867,
        -0.0007492331787943847,
        0.0022998321801424026,
        0.00014541856944561005
      ],
      "p_value": 0.57535,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "Exact generator derivative channels should improve learning of derivative-dependent targets relative to finite differences.",
      "observed": {
        "baseline_best_mse": 0.01434623496606946,
        "idea_best_mse": 0.014594940701499581
      },
      "confirmed": false
    },
    "custom_track": {
      "name": "so3_spectral_regression",
      "file": "so3_custom_track.py",
      "domain": "SO(3) orientation fields"
    }
  },
  "system_verdict": "failed",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 0,
  "system_judged": true
}