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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Implemented and evaluated an end-to-end quadratic bottleneck on the structurally matched controlled-dynamics track, using paired GRU systems with identical data, training budget, and learning-rate search space. The analytic tangent sanity check passed, and the trained quadratic contribution scaled with the predicted slope of 2.0, confirming the mechanism quantitatively. However, the quadratic system had higher mean test MSE than the baseline (0.000373633 vs 0.000306926), with paired delta +0.0000667074 and permutation p=0.6974, so there was no significant improvement.",
  "metrics": {
    "baseline": "Best lr=0.01; full 8-seed test MSE mean 0.000306925562, std 0.000114631221",
    "idea": "Best lr=0.01; full 8-seed test MSE mean 0.000373633002, std 0.000292045128; paired delta +0.0000667074401; p=0.6974; 5/8 paired wins",
    "mechanism_signature": "Trained quadratic decoder output contribution norm versus latent scaling had observed log-log slope 1.999999995 versus predicted 2.0; confirmed=true; parameters baseline=13777, idea=13840"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 quadratic_bench.py",
  "files": [
    "quadratic_bench.py",
    "bench_report.json"
  ],
  "limitations": "Only the dynamics track was tested; long-horizon rollouts, explicit invariance-loss training, held-out forcing-frequency tracking, and larger latent-dimension sweeps were not tested. The implementation tests an explicit quadratic readout bottleneck end-to-end but does not add the full neural invariance residual loss because the built-in benchmark exposes a supervised one-step prediction task rather than a vector-field training interface.",
  "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": 0.0021582036570180207
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.0010473794682184234
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.0003684468611027114
        }
      ],
      "full": {
        "mean": 0.00030692556174471974,
        "std": 0.00011463122148079786,
        "per_seed": [
          0.00029358387109823525,
          0.0005360718932934105,
          0.0002490969200152904,
          0.00039503476000390947,
          0.00015615242591593415,
          0.00021396420197561383,
          0.00037629096186719835,
          0.000235209459788166
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 0.00037363300180004444,
      "std": 0.00029204512808802147,
      "per_seed": [
        0.0005500055267475545,
        0.0004540196096058935,
        0.0009904648177325726,
        0.0003427359333727509,
        6.876005500089377e-05,
        4.5171487727202475e-05,
        0.0004113739705644548,
        0.000126532613649033
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 6.67074400553247e-05,
      "idea_wins": 5,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.0002564216556493193,
        -8.205228368751705e-05,
        0.0007413678977172822,
        -5.229882761185859e-05,
        -8.739237091504037e-05,
        -0.00016879271424841136,
        3.5083008697256446e-05,
        -0.00010867684613913298
      ],
      "p_value": 0.6974,
      "mde": 0.0002525162608655287,
      "mde_rel_pct": 82.27280237921497,
      "verdict": "no significant win",
      "system_worked": false
    },
    "mechanism_signature": {
      "math_check": {
        "central_difference_max_errors": [
          1.5987211554602254e-14,
          3.108624468950438e-14,
          1.5631940186722204e-13,
          5.53335155473178e-13
        ],
        "quadratic_value_scale_slope": 2.0000000000000018,
        "prediction": {
          "finite_difference": "zero up to roundoff",
          "value_scale": 2.0
        },
        "passed": true
      },
      "trained_model": {
        "quantity": "trained quadratic decoder output contribution norm",
        "alphas": [
          0.25,
          0.5,
          1.0,
          2.0
        ],
        "observed_norms": [
          0.052310265600681305,
          0.20924106240272522,
          0.8369642496109009,
          3.3478569984436035
        ],
        "predicted_loglog_slope": 2.0,
        "observed_loglog_slope": 1.9999999945043294,
        "confirmed": true,
        "baseline_parameters": 13777,
        "idea_parameters": 13840
      },
      "idea_sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.002017721548327245,
          "std": 0.0007681260086240184
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.0009736494830576703,
          "std": 0.00018790902759927578
        },
        {
          "cfg": {
            "lr": 0.01
          },
          "mean": 0.00037363300180004444,
          "std": 0.00029204512808802147
        }
      ],
      "protocol_note": "Dynamics selected because the idea targets controlled state evolution; paired GRU base and latent bottleneck are identical."
    }
  },
  "system_verdict": "failed",
  "practical_verdict": "inconclusive",
  "mechanism_ok": 0,
  "system_judged": true
}