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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "The full structurally matched benchmark was completed on the registered translated_density_transport track with independently trained baseline and monotone CDT systems, identical MLP capacity, shared learning-rate grid, and eight paired seeds. The monotone mechanism was confirmed quantitatively, but task MSE was significantly worse for the idea: 4.1069303155 versus 0.0046020751 for the baseline, delta +4.1023282403, permutation p=0.0081.",
  "metrics": {
    "baseline": "Best lr=0.009; full 8-seed test MSE mean 0.004602075146976858.",
    "idea": "Best lr=0.009; full 8-seed test MSE mean 4.106930315494537; monotonic fraction 1.0; negative fraction 0.0; mass proxy error 2.808868893033889e-08."
  },
  "bench_report": {
    "bench_version": 1,
    "track": "translated_density_transport",
    "model": "shared_mlp",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.009
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 0.6033583730459213
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 0.06099952198565006
        },
        {
          "cfg": {
            "lr": 0.009
          },
          "mean": 0.005193391698412597
        }
      ],
      "full": {
        "mean": 0.004602075146976858,
        "std": 0.0010999913497623144,
        "per_seed": [
          0.004353613592684269,
          0.00704985624179244,
          0.005014418624341488,
          0.0043556783348321915,
          0.004588612355291843,
          0.0034741132985800505,
          0.004819120746105909,
          0.003161187982186675
        ],
        "n": 8
      }
    },
    "idea": {
      "mean": 4.106930315494537,
      "std": 0.0030807169256581463,
      "per_seed": [
        4.1090593338012695,
        4.107220649719238,
        4.104761600494385,
        4.1017680168151855,
        4.108211040496826,
        4.103658676147461,
        4.111758232116699,
        4.109004974365234
      ],
      "n": 8,
      "best_cfg": {
        "lr": 0.009
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 4.110945641994476,
          "std": 0.003565288253498607,
          "per_seed": [
            4.113897800445557,
            4.113440036773682,
            4.10503101348877,
            4.1087565422058105,
            4.1114397048950195,
            4.106204509735107,
            4.114856719970703,
            4.113938808441162
          ],
          "n": 8
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 4.108930468559265,
          "std": 0.002267987204719172,
          "per_seed": [
            4.110673427581787,
            4.107449531555176,
            4.107608795166016,
            4.105695724487305,
            4.110192775726318,
            4.106705188751221,
            4.112852096557617,
            4.110266208648682
          ],
          "n": 8
        },
        {
          "cfg": {
            "lr": 0.009
          },
          "mean": 4.106930315494537,
          "std": 0.0030807169256581463,
          "per_seed": [
            4.1090593338012695,
            4.107220649719238,
            4.104761600494385,
            4.1017680168151855,
            4.108211040496826,
            4.103658676147461,
            4.111758232116699,
            4.109004974365234
          ],
          "n": 8
        }
      ]
    },
    "comparison": {
      "delta_mean": 4.1023282403475605,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        4.104705720208585,
        4.100170793477446,
        4.099747181870043,
        4.097412338480353,
        4.103622428141534,
        4.100184562848881,
        4.106939111370593,
        4.105843786383048
      ],
      "p_value": 0.0081,
      "mde": 0.0028447563556861727,
      "mde_rel_pct": 61.81464371686579,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "trained_models": true,
      "prediction": "translated states should produce monotone quantiles with approximately constant +0.08 displacement, nonnegative decoded densities, and unit mass",
      "predicted_vs_observed": {
        "expected_quantile_shift": 0.08,
        "observed_mean_quantile_shift": 0.05113657424226403,
        "observed_monotone_fraction": 1.0,
        "observed_idea_negative_fraction": 0.0,
        "observed_idea_mean_mass_proxy_error": 2.808868893033889e-08,
        "observed_baseline_negative_fraction": 0.3139062523841858
      },
      "confirmed": true,
      "paired_values": [
        {
          "seed": 0,
          "baseline": 0.004353613592684269,
          "idea": 4.1090593338012695
        },
        {
          "seed": 1,
          "baseline": 0.00704985624179244,
          "idea": 4.107220649719238
        },
        {
          "seed": 2,
          "baseline": 0.005014418624341488,
          "idea": 4.104761600494385
        },
        {
          "seed": 3,
          "baseline": 0.0043556783348321915,
          "idea": 4.1017680168151855
        },
        {
          "seed": 4,
          "baseline": 0.004588612355291843,
          "idea": 4.108211040496826
        },
        {
          "seed": 5,
          "baseline": 0.0034741132985800505,
          "idea": 4.103658676147461
        },
        {
          "seed": 6,
          "baseline": 0.004819120746105909,
          "idea": 4.111758232116699
        },
        {
          "seed": 7,
          "baseline": 0.003161187982186675,
          "idea": 4.109004974365234
        }
      ],
      "idea_sweep": [
        {
          "cfg": {
            "lr": 0.001
          },
          "mean": 4.110945641994476,
          "std": 0.003565288253498607,
          "per_seed": [
            4.113897800445557,
            4.113440036773682,
            4.10503101348877,
            4.1087565422058105,
            4.1114397048950195,
            4.106204509735107,
            4.114856719970703,
            4.113938808441162
          ],
          "n": 8
        },
        {
          "cfg": {
            "lr": 0.003
          },
          "mean": 4.108930468559265,
          "std": 0.002267987204719172,
          "per_seed": [
            4.110673427581787,
            4.107449531555176,
            4.107608795166016,
            4.105695724487305,
            4.110192775726318,
            4.106705188751221,
            4.112852096557617,
            4.110266208648682
          ],
          "n": 8
        },
        {
          "cfg": {
            "lr": 0.009
          },
          "mean": 4.106930315494537,
          "std": 0.0030807169256581463,
          "per_seed": [
            4.1090593338012695,
            4.107220649719238,
            4.104761600494385,
            4.1017680168151855,
            4.108211040496826,
            4.103658676147461,
            4.111758232116699,
            4.109004974365234
          ],
          "n": 8
        }
      ]
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_transport_bench.py",
  "files": [
    "stage2_transport_bench.py",
    "custom_transport_track.py",
    "bench_report.json",
    "bench_stdout.txt"
  ],
  "limitations": "Only the registered 1D translated-density transport track was tested. Higher-dimensional fields, POD bottlenecks, learned latent dimensions d=2/4/8, alternative push-forward discretizations, and longer training budgets were not evaluated.",
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}