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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "The WOS stochastic-target system was significantly worse than the tuned exact-label baseline on the structurally matched PDE benchmark. The sphere identity and inverse-K variance check passed, but the required task metric comparison did not show a win.",
  "metrics": {
    "baseline": "Mean test MSE 0.0022902884520590305; tuned lr=0.01.",
    "idea": "Mean test MSE 0.01020078727742657; best lr=0.003, K=8, M=24."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py",
  "files": [
    "stage2_bench.py",
    "bench_report.json",
    "poisson_track.py"
  ],
  "limitations": "Only registered poisson_boundary and mlp_tiny were tested; no higher-dimensional or screened-Poisson task, FLOP timing, or true residual PINN baseline.",
  "bench_report": {
    "bench_version": 1,
    "track": "poisson_boundary",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.046625981107354164
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.005821306491270661
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.0026237370329909027
        }
      ],
      "full": {
        "mean": 0.0022902884520590305,
        "std": 0.0008475947478694112,
        "per_seed": [
          0.002140381373465061,
          0.0017085936851799488,
          0.0028188657015562057,
          0.003827107371762395,
          0.0007765397895127535,
          0.0018183346837759018,
          0.002519753761589527,
          0.002712731249630451
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 0.01020078727742657,
      "std": 0.00296426428649209,
      "per_seed": [
        0.009175088256597519,
        0.007033985573798418,
        0.006878904066979885,
        0.015077905729413033,
        0.014966735616326332,
        0.009611713699996471,
        0.009160290472209454,
        0.009701674804091454
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 0.00791049882536754,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.007034706883132458,
        0.005325391888618469,
        0.004060038365423679,
        0.011250798357650638,
        0.014190195826813579,
        0.00779337901622057,
        0.0066405367106199265,
        0.006988943554461002
      ],
      "p_value": 0.0081,
      "mde": 0.0027432218817820646,
      "mde_rel_pct": 119.77626134017463,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "WOS target variance decreases approximately as 1/K",
      "predicted_ratio": 1.0,
      "observed_ratio": 0.899,
      "K_product_cv": 0.03766105478867267,
      "confirmed": true
    },
    "custom_track": {
      "name": "poisson_boundary",
      "file": "bench/custom_tracks/poisson_boundary.py",
      "domain": "pde"
    },
    "structural_match": "Registered PDE Dirichlet boundary-value regression on the unit disk; WOS is the intervention."
  },
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}