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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Ran the full 8-seed paired benchmark on the registered two_view_gauge_localization_v2 track using identical mlp_tiny systems and a tuned baseline sweep. The coarse-to-fine variant had higher test MSE (1.1728449 vs 1.1499638), paired delta +0.0228810, and permutation p=0.12625, so there is no significant win. The trained-model mechanism signature was also not confirmed: refinement slightly increased MAE from 0.9011721 to 0.9014216.",
  "metrics": {
    "baseline": "Best lr=0.01; full 8-seed MSE mean 1.1499638408 ± 0.1032812345",
    "idea": "lr=0.01; full 8-seed MSE mean 1.1728448644 ± 0.1088541057; paired delta +0.0228810236; p=0.12625"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 registered_bench.py",
  "files": [
    "registered_bench.py",
    "registered_bench_report.json",
    "registered_bench_stdout.json"
  ],
  "limitations": "The registered localization track predicts a scalar coordinate rather than a variable number of sources, so it only partially exercises the proposed neural slot-head use case. The intervention is a differentiable soft coarse-grid contraction, not the full learned-subspace projection-residual score with analytic Fourier gradient, fixed kappa^-2 refinement, NMS, and runtime evaluation accounting. Baseline sweep used three shared learning rates and the idea included the best plus two nearby settings; no larger architecture or additional localization tracks were tested.",
  "bench_report": {
    "bench_version": 1,
    "track": "two_view_gauge_localization_v2",
    "model": "mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 1.2723695039749146
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 1.2335557341575623
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 1.0947982370853424
        }
      ],
      "full": {
        "mean": 1.149963840842247,
        "std": 0.1032812344962866,
        "per_seed": [
          1.0112413167953491,
          1.0117237567901611,
          1.1340291500091553,
          1.222198724746704,
          1.1495081186294556,
          1.266226053237915,
          1.3084473609924316,
          1.0963362455368042
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 1.1728448644280434,
      "std": 0.10885410569769803,
      "per_seed": [
        0.9953804612159729,
        1.0588462352752686,
        1.2159733772277832,
        1.2204293012619019,
        1.1976563930511475,
        1.2475311756134033,
        1.355494499206543,
        1.0914474725723267
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 0.022881023585796356,
      "idea_wins": 4,
      "n_pairs": 8,
      "per_seed_diffs": [
        -0.01586085557937622,
        0.04712247848510742,
        0.08194422721862793,
        -0.001769423484802246,
        0.048148274421691895,
        -0.01869487762451172,
        0.04704713821411133,
        -0.004888772964477539
      ],
      "p_value": 0.12625,
      "mde": 0.03145608783128045,
      "mde_rel_pct": 2.7353979937527115,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "coarse proposal plus fixed refinement should reduce raw localization error",
      "raw_mae_mean": 0.9011721238493919,
      "refined_mae_mean": 0.9014216214418411,
      "mean_proposal_shift": 0.017849167110398412,
      "confirmed": false
    },
    "idea_nearby_sweep": {
      "0.001": {
        "mean": 1.2723439633846283,
        "std": 0.07400079392237602,
        "per_seed": [
          1.1923704147338867,
          1.2113070487976074,
          1.3731907606124878,
          1.3125076293945312
        ],
        "n": 4
      },
      "0.01": {
        "mean": 1.1226573437452316,
        "std": 0.098156125855815,
        "per_seed": [
          0.9953804612159729,
          1.0588462352752686,
          1.2159733772277832,
          1.2204293012619019
        ],
        "n": 4
      }
    },
    "custom_track": {
      "name": "two_view_gauge_localization_v2",
      "registered": true,
      "domain": "geometry/localization"
    }
  },
  "system_verdict": "failed",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 0,
  "system_judged": true
}