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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "The registered built-in sequence track was benchmarked with 8 paired seeds, a tuned baseline sweep, and a parity-matched idea learning-rate sweep. Sinkhorn mixing confirmed the trained-model mechanism prediction, but the idea was significantly worse on independent sequence test MSE: 0.187322 versus 0.162098, paired delta +0.025224, permutation p=0.0081.",
  "metrics": {
    "baseline": "transformer_tiny stream baseline, best lr=0.0015, test MSE 0.162098 ± 0.024682",
    "idea": "Sinkhorn doubly-stochastic stream transformer, best lr=0.002, test MSE 0.187322 ± 0.025152; 0/8 paired wins"
  },
  "bench_report": {
    "bench_version": 1,
    "track": "sequence",
    "model": "transformer_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.0015
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.0007
          },
          "mean": 0.18085692822933197
        },
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.17446092143654823
        },
        {
          "cfg": {
            "lr": 0.0015
          },
          "mean": 0.17426595836877823
        },
        {
          "cfg": {
            "lr": 0.002
          },
          "mean": 0.1786409355700016
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.18349025771021843
        }
      ],
      "full": {
        "mean": 0.16209838539361954,
        "std": 0.024681871508716122,
        "per_seed": [
          0.1827893853187561,
          0.13988947868347168,
          0.18467898666858673,
          0.1897059828042984,
          0.11963248252868652,
          0.15095271170139313,
          0.18380902707576752,
          0.14532902836799622
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 0.18732197023928165,
      "std": 0.025151544057018065,
      "per_seed": [
        0.19494226574897766,
        0.15557610988616943,
        0.19609080255031586,
        0.2065867781639099,
        0.1564088761806488,
        0.16889484226703644,
        0.23473307490348816,
        0.18534301221370697
      ],
      "n": 8
    },
    "comparison": {
      "delta_mean": 0.025223584845662117,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.012152880430221558,
        0.015686631202697754,
        0.011411815881729126,
        0.01688079535961151,
        0.03677639365196228,
        0.01794213056564331,
        0.05092404782772064,
        0.040013983845710754
      ],
      "p_value": 0.0081,
      "mde": 0.01259364706714858,
      "mde_rel_pct": 7.769137882877572,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "trained_model": true,
      "predicted": {
        "max_frobenius_ratio": "<=1",
        "row_col_error": "0"
      },
      "observed": {
        "max_frobenius_ratio": 0.7491575479507446,
        "mean_frobenius_ratio": 0.748988151550293,
        "max_row_error": 2.384185791015625e-07,
        "max_col_error": 1.1920928955078125e-07
      },
      "confirmed": true
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py",
  "files": [
    "stage2_bench.py",
    "bench_report.json"
  ],
  "limitations": "Only the structurally matched sequence forecast track was tested; tabular, vision, and dynamics were not tested. The benchmark used 12 epochs and 800 training/300 test examples rather than full dataset sizes, and did not measure FLOPs, wall-clock speed, gradient spikes, or long-depth sensitivity.",
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}