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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "The entropy-calibrated curvature intervention was tested on the registered dynamics track with the shared rnn_small architecture, 8 paired seeds, and a parity-preserving learning-rate sweep. It improved mean test MSE from 0.00182823 to 0.00141185 and won 7/8 pairs, but the permutation p-value was 0.05505, just above 0.05; therefore this is not a significant win.",
  "metrics": {
    "baseline": "Best tuned fixed baseline lr=0.006, 12 epochs; mean test MSE=0.0018282315868418664.",
    "idea": "Best entropy-curvature setting lr=0.006, 12 epochs; mean test MSE=0.0014118490216787905; paired delta=-0.0005000870587537065; 7/8 wins; p=0.05505."
  },
  "bench_report": {
    "bench_version": 1,
    "track": "dynamics",
    "model": "rnn_small",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_config": {
        "lr": 0.006,
        "epochs": 12
      },
      "sweep": {
        "0.001": 0.004298779065720737,
        "0.003": 0.0034615940821822733,
        "0.006": 0.0018282315868418664
      },
      "full": {
        "per_seed": [
          0.002004395006224513,
          0.001437882543541491,
          0.0015748648438602686,
          0.002295783953741193,
          0.0024020641576498747,
          0.0024898042902350426,
          0.0020851334556937218,
          0.0025566271506249905
        ]
      },
      "all_configs": {
        "0.001": [
          0.007754037156701088,
          0.003745198715478182,
          0.0025475025177001953,
          0.003148377873003483,
          0.004954223055392504,
          0.004425588063895702,
          0.0044244141317903996,
          0.009162315167486668
        ],
        "0.003": [
          0.002470282604917884,
          0.002755755092948675,
          0.0011578478151932359,
          0.007462490815669298,
          0.0025412666145712137,
          0.0030948929488658905,
          0.0021244161762297153,
          0.0019843443296849728
        ],
        "0.006": [
          0.002004395006224513,
          0.001437882543541491,
          0.0015748648438602686,
          0.002295783953741193,
          0.0024020641576498747,
          0.0024898042902350426,
          0.0020851334556937218,
          0.0025566271506249905
        ]
      }
    },
    "idea": {
      "best_config": {
        "lr": 0.006,
        "epochs": 12
      },
      "sweep_means": {
        "0.001": 0.0045381716336123645,
        "0.003": 0.0035767197841778398,
        "0.006": 0.0014118490216787905
      },
      "per_seed": [
        0.0018144856439903378,
        0.0014102724380791187,
        0.0007331487722694874,
        0.0016894892323762178,
        0.0015310372691601515,
        0.0012987027876079082,
        0.002770845778286457,
        0.0015978770097717643
      ],
      "signature_samples": [
        {
          "entropy": 0.7434267299673585,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.352136492729187,
          "radius_tau": 0.3528751730918884
        },
        {
          "entropy": 0.8414160177487802,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.300600290298462,
          "radius_tau": 0.3686612844467163
        },
        {
          "entropy": 0.8015473816323408,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.31516432762146,
          "radius_tau": 0.41591450572013855
        },
        {
          "entropy": 0.7520167144600518,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.4210695028305054,
          "radius_tau": 0.40408262610435486
        },
        {
          "entropy": 0.8041456100108358,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.2326090335845947,
          "radius_tau": 0.40702223777770996
        },
        {
          "entropy": 0.7911075937989723,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.3452383279800415,
          "radius_tau": 0.3651469051837921
        },
        {
          "entropy": 0.8041456100108358,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.3425686359405518,
          "radius_tau": 0.41605350375175476
        },
        {
          "entropy": 0.7550113436280443,
          "curvature": 0.056044551169079165,
          "radius_mean": 1.4108587503433228,
          "radius_tau": 0.4241693317890167
        }
      ]
    },
    "comparison": {
      "delta_mean": -0.0005000870587537065,
      "idea_wins": 7,
      "n_pairs": 8,
      "per_seed_diffs": [
        -0.0001899093622341752,
        -2.7610105462372303e-05,
        -0.0008417160715907812,
        -0.000606294721364975,
        -0.0008710268884897232,
        -0.0011911015026271343,
        0.0006857123225927353,
        -0.0009587501408532262
      ],
      "p_value": 0.05505,
      "mde": 0.0005172797403971514,
      "mde_rel_pct": 24.56429711910889,
      "verdict": "no significant win",
      "system_worked": false
    },
    "math_sanity": {
      "entropy_decreasing_fraction": 1.0,
      "lambda_true": 1.1249358995152312,
      "lambda_recovered": 1.1249358995152312,
      "amplification": 1.657187089473768
    },
    "mechanism_signature": {
      "predicted_lambda": 1.1249358995152312,
      "observed_entropy_curve_recovered_lambda": 1.1249358995152312,
      "trained_model_entropy_mean": 0.7866021251571524,
      "trained_model_curvature_mean": 0.056044551169079165,
      "confirmed": true
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 bench_entropy_curvature.py",
  "files": [
    "bench_entropy_curvature.py",
    "bench_report.json",
    "bench_run.log"
  ],
  "limitations": "Only the registered dynamics track was tested. The implementation uses a Poincare-like distance proxy and curvature-dependent radial regularization rather than a full Poincare-ball encoder and exact hyperbolic prototype classifier. The observed improvement narrowly missed the required p<0.05 significance threshold.",
  "system_verdict": "partial",
  "practical_verdict": "no_effect",
  "mechanism_ok": 1,
  "system_judged": true
}