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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "Built and evaluated TD-to-PDE continuation on the registered poisson_boundary PDE track using a shared smooth MLP. Baseline test MSE was 0.0197138, while the best continuation setting achieved 0.0868471; paired delta was +0.0671333 with permutation p=0.0081, so the idea was significantly worse. The trained-model mechanism signature was also not confirmed because PDE residual increased from 0.0057077 early to 0.7555173 late.",
  "metrics": {
    "baseline": "poisson_boundary, smooth_mlp_shared, tuned lr=0.01; 8-seed mean test MSE 0.019713770831003785",
    "idea": "poisson_boundary, smooth_mlp_shared, continuation lr=0.003 and PDE weight 1.0; 8-seed mean test MSE 0.0868471022695303"
  },
  "bench_report": {
    "bench_version": 1,
    "track": "poisson_boundary",
    "model": "smooth_mlp_shared",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.09443562291562557
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.026866592932492495
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.021620961604639888
        }
      ],
      "best_cfg": {
        "lr": 0.01
      },
      "full": {
        "config": {
          "lr": 0.01,
          "pde_weight": 0.0
        },
        "per_seed": [
          0.027827076613903046,
          0.013140390627086163,
          0.02447698265314102,
          0.02103939652442932,
          0.019167248159646988,
          0.018314614295959472,
          0.017920590192079543,
          0.017704755663871766
        ],
        "mean": 0.019713770831003785,
        "std": 0.004566942238936102
      }
    },
    "idea": {
      "config": {
        "lr": 0.003,
        "pde_weight": 1.0
      },
      "per_seed": [
        0.10768520832061768,
        0.07102982699871063,
        0.07831890136003494,
        0.08951634913682938,
        0.08078285306692123,
        0.07756161689758301,
        0.10001540929079056,
        0.08986665308475494
      ],
      "mean": 0.0868471022695303,
      "std": 0.011563873184926184
    },
    "comparison": {
      "delta_mean": 0.06713333143852651,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.07985813170671463,
        0.05788943637162447,
        0.05384191870689392,
        0.06847695261240005,
        0.061615604907274246,
        0.06332949828356504,
        0.08203655108809471,
        0.07001855783164501
      ],
      "p_value": 0.0081,
      "mde": 0.008368594524021243,
      "mde_rel_pct": 42.45050120426469,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "predictive loss stabilizes before strong PDE reduction",
      "early_predictive_loss": 0.1310851002732913,
      "late_predictive_loss": 0.09046961230171077,
      "early_pde_residual": 0.005707668307877611,
      "late_pde_residual": 0.7555173069182167,
      "confirmed": false,
      "math_check_max_abs_laplacian": 0.0
    },
    "custom_track": {
      "name": "poisson_boundary",
      "file": "/home/maxwelhelp/all/math2nn/bench/custom_tracks/poisson_boundary.py",
      "domain": "pde"
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 run_bench.py",
  "files": [
    "poisson_track.py",
    "run_bench.py",
    "bench_report.json"
  ],
  "limitations": "Only the registered poisson_boundary PDE track and one shared smooth MLP architecture were tested. Genuine stochastic TD transitions, actor-critic training, action maximization, alternative PDE weights beyond the tested continuation sweep, and larger models were not evaluated.",
  "system_verdict": "failed",
  "practical_verdict": "harms",
  "mechanism_ok": 0,
  "system_judged": true
}