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

(универсальный)

Ответ агента:

{
  "worked": true,
  "confidence": 8,
  "verdict": "Built a structurally matched custom point-cloud benchmark because no built-in track represents tangent-surface geometry. Across 8 paired seeds, the independently trained tangent-projected system reduced MSE from 0.027478 to 0.021507, with delta -0.005971 and permutation p=0.03805, meeting the harness criterion for a significant win. The trained-model mechanism signature confirmed normal leakage suppression (0.6703 raw mean component versus 4.77e-7 projected maximum), but rotation prediction change was 25.5%, so full rotational invariance was not achieved.",
  "metrics": {
    "baseline": "MSE 0.0274779503; best lr=0.01; 8-seed mean",
    "idea": "MSE 0.0215066459; best lr=0.01; 8-seed mean; absolute improvement 0.0059713044 (21.73%)",
    "paired_delta": "-0.0059713044",
    "permutation_p": "0.03805",
    "idea_wins": "7/8",
    "mde": "0.0053456830 (19.45% of baseline mean)"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 run_stage2.py",
  "files": [
    "tangent_surface_track.py",
    "run_stage2.py",
    "stage2_report.json"
  ],
  "limitations": "This is a small synthetic sphere-based custom track with 400 training and 100 test samples per seed, a shared MLP rather than a full multi-layer message-passing network, 14 epochs, and three learning-rate settings. It tests tangent projection and normal suppression but not torus/deformed-isometric transfer, weighted least-squares Jacobian features, FLOP matching beyond the shared MLP shape, large-scale point clouds, or the proposed <1% rotation-change target; measured rotation prediction change was 25.5%.",
  "bench_report": {
    "bench_version": 1,
    "track": "tangent_surface_vector_regression",
    "model": "shared_mlp",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.03363563574284315
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.030874531697183846
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.028123383371797372
        }
      ],
      "full": {
        "mean": 0.02747795032337308,
        "std": 0.006638132620035498,
        "per_seed": [
          0.02969777025282383,
          0.03375587304379046,
          0.02329469621181488,
          0.024466974660754204,
          0.024169445037841797,
          0.027743976563215256,
          0.029445959255695344,
          0.027241207510590552
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 0.021506645949557424,
      "std": 0.005512922742672397,
      "per_seed": [
        0.018548764288425446,
        0.026988901843085885,
        0.0270770076662302,
        0.02075173147022724,
        0.007078237831294537,
        0.02547464147210121,
        0.027338292449712753,
        0.018424329161643983
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": -0.005971304373815656,
      "idea_wins": 7,
      "n_pairs": 8,
      "per_seed_diffs": [
        -0.011149005964398384,
        -0.006766971200704575,
        0.0037823114544153214,
        -0.0037152431905269623,
        -0.017091207206249237,
        -0.002269335091114044,
        -0.0021076668053865433,
        -0.008453316986560822
      ],
      "p_value": 0.03805,
      "mde": 0.005345683018986985,
      "mde_rel_pct": 19.45444604155894,
      "verdict": "idea better (significant)",
      "system_worked": true
    },
    "mechanism_signature": {
      "raw_mean_normal_component": 0.6702553331851959,
      "projected_max_normal_component": 4.76837158203125e-07,
      "rotation_relative_prediction_change": 0.25490777008235455,
      "predicted_projected_leakage": "<1e-5",
      "confirmed": true
    },
    "custom_track": {
      "name": "tangent_surface_vector_regression",
      "file": "tangent_surface_track.py",
      "domain": "geometry/point-cloud"
    }
  },
  "system_verdict": "worked",
  "practical_verdict": "helps",
  "mechanism_ok": 1,
  "system_judged": true
}