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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "The spectrally identifiable phaseless recurrent layer was implemented and evaluated on the registered dynamics track. Its phase-invariance mechanism was confirmed, but it was significantly worse on task MSE than the tuned GRU baseline: 0.0180801 versus 0.000393844, paired delta +0.0176862 with permutation p=0.0081.",
  "metrics": {
    "baseline": "Mean test MSE 0.0003938444 over 8 seeds; best learning rate 0.01.",
    "idea": "Mean test MSE 0.01808006 over 8 seeds; best tested learning rate 0.01; 0/8 paired wins."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py",
  "files": [
    "stage2_bench.py",
    "bench_report.json"
  ],
  "limitations": "The idea model has more parameters than the GRU baseline (24961 vs 13313). It mean-pools the input window before spectral encoding rather than evolving a spectral hidden state at every input step. Larger datasets, alternative graphs, sensor noise, longer training, and a K sweep were not tested.",
  "bench_report": {
    "bench_version": 1,
    "track": "dynamics",
    "model": "rnn_small",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.0014828930143266916
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.000868373506818898
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.0004153431436861865
        }
      ],
      "full": {
        "mean": 0.0003938443751394516,
        "std": 0.00015694612826889552,
        "per_seed": [
          0.0004267372714821249,
          0.00025510918931104243,
          0.00029441321385093033,
          0.0006851129001006484,
          0.00039094407111406326,
          0.0005367420380935073,
          0.00014661152090411633,
          0.0004150847962591797
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 0.01808006438659504,
      "std": 0.006827547286877937,
      "per_seed": [
        0.015054022893309593,
        0.01623576134443283,
        0.03221019357442856,
        0.00664726784452796,
        0.021471986547112465,
        0.021092308685183525,
        0.016446856781840324,
        0.015482117421925068
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 0.01768622001145559,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.014627285621827468,
        0.01598065215512179,
        0.03191578036057763,
        0.005962154944527311,
        0.0210810424759984,
        0.020555566647090018,
        0.016300245260936208,
        0.015067032625665888
      ],
      "p_value": 0.0081,
      "mde": 0.006166672867800231,
      "mde_rel_pct": 1565.763854216973,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": "global phase is unobservable by magnitude trajectory",
      "predicted_max_abs_difference": 0.0,
      "observed_mean_abs_difference": 1.0522416005187551e-07,
      "operator_pair_sum_gap": 6.656185635112166e-06,
      "operator_sigma_min_S": 0.009892589246225832,
      "confirmed": true
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.01
        },
        "result": {
          "mean": 0.01808006438659504,
          "std": 0.006827547286877937,
          "per_seed": [
            0.015054022893309593,
            0.01623576134443283,
            0.03221019357442856,
            0.00664726784452796,
            0.021471986547112465,
            0.021092308685183525,
            0.016446856781840324,
            0.015482117421925068
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.001
        },
        "result": {
          "mean": 0.01906815404072404,
          "std": 0.004254108046330375,
          "per_seed": [
            0.018856093287467957,
            0.01202596165239811,
            0.0262176301330328,
            0.024596311151981354,
            0.019076775759458542,
            0.01587844267487526,
            0.017323212698101997,
            0.018570804968476295
          ],
          "n": 8
        }
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "result": {
          "mean": 0.018212101538665593,
          "std": 0.0021132105733096982,
          "per_seed": [
            0.017741071060299873,
            0.01399772334843874,
            0.02053205855190754,
            0.0214164387434721,
            0.018905995413661003,
            0.018360335379838943,
            0.017668212698101995,
            0.017074882984161377
          ],
          "n": 8
        }
      }
    ],
    "parameter_counts": {
      "baseline": 13313,
      "idea": 24961
    }
  },
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}