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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 10,
  "verdict": "Built an end-to-end robust relative-degree-2 CBF projection after a neural pendulum policy, including actuator bounds, disturbance-margin tightening, and emergency fallback. The baseline mean test MSE was 0.00199788 versus 0.05749530 for the best robust-CBF setting; paired delta was +0.05549742 with permutation p=0.0081, so the idea was significantly worse. The mechanism signature nevertheless confirmed the predicted margin slope (-2.0) and lower-bound shift (+1.0 per epsilon) from trained models.",
  "metrics": {
    "baseline": "mean MSE=0.0019978795608039945, best lr=0.006, epochs=15",
    "idea": "mean MSE=0.05749530019238591, best lr=0.003, epsilon=0.25, epochs=15; delta=+0.05549742063158192; p=0.0081; 0/8 paired wins"
  },
  "bench_report": {
    "bench_version": 1,
    "track": "robust_cbf_pendulum_policy",
    "model": "local_mlp_tiny",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.006,
        "eps": 0.0,
        "epochs": 15
      },
      "sweep": [
        {
          "cfg": {
            "lr": 0.001,
            "eps": 0.0,
            "epochs": 15
          },
          "mean": 0.021794200874865055
        },
        {
          "cfg": {
            "lr": 0.003,
            "eps": 0.0,
            "epochs": 15
          },
          "mean": 0.002334954886464402
        },
        {
          "cfg": {
            "lr": 0.006,
            "eps": 0.0,
            "epochs": 15
          },
          "mean": 0.0018331437022425234
        }
      ],
      "full": {
        "mean": 0.0019978795608039945,
        "std": 0.0005495084055684182,
        "per_seed": [
          0.0015628907131031156,
          0.0015461408765986562,
          0.0021467802580446005,
          0.0020767629612237215,
          0.0030877823010087013,
          0.0011752876453101635,
          0.0020312685519456863,
          0.0023561231791973114
        ],
        "n": 8
      }
    },
    "idea": {
      "config": {
        "lr": 0.003,
        "eps": 0.25,
        "epochs": 15
      },
      "mean": 0.05749530019238591,
      "std": 0.00835564815799482,
      "per_seed": [
        0.06218996271491051,
        0.06785300374031067,
        0.04927119240164757,
        0.05358738824725151,
        0.042678408324718475,
        0.06384517252445221,
        0.05430787801742554,
        0.06622939556837082
      ],
      "n": 8,
      "selection_sweep": [
        {
          "cfg": {
            "lr": 0.001,
            "eps": 0.15,
            "epochs": 15
          },
          "mean": 0.05939904507249594
        },
        {
          "cfg": {
            "lr": 0.003,
            "eps": 0.25,
            "epochs": 15
          },
          "mean": 0.058225386776030064
        },
        {
          "cfg": {
            "lr": 0.006,
            "eps": 0.35,
            "epochs": 15
          },
          "mean": 0.06890939734876156
        }
      ]
    },
    "comparison": {
      "delta_mean": 0.05549742063158192,
      "idea_wins": 0,
      "n_pairs": 8,
      "per_seed_diffs": [
        0.06062707200180739,
        0.06630686286371201,
        0.04712441214360297,
        0.05151062528602779,
        0.039590626023709774,
        0.06266988487914205,
        0.05227660946547985,
        0.06387327238917351
      ],
      "p_value": 0.0081,
      "mde": 0.007832081629823283,
      "mde_rel_pct": 392.0197084688862,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "prediction": {
        "margin_slope": -2.0,
        "lower_bound_shift_per_epsilon": 1.0
      },
      "observed_from_trained_models": {
        "margin_slope_mean": -2.0,
        "lower_shift_mean": 1.0,
        "mean_intervention": 1.705237858260225,
        "feasible_fraction_mean": 1.0,
        "n_models": 8
      },
      "confirmed": true
    },
    "custom_track": {
      "name": "robust_cbf_pendulum_policy",
      "file": "robust_cbf_bench.py",
      "domain": "dynamics"
    }
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 robust_cbf_bench.py",
  "files": [
    "robust_cbf_bench.py",
    "bench_report.json"
  ],
  "limitations": "The custom benchmark is supervised state-to-expert-action regression rather than closed-loop reinforcement learning. It uses a known two-state pendulum surrogate, fixed disturbance-bound settings, scalar action, and does not measure runtime, multi-input quadratic programming, learned model uncertainty, observer estimation error, or long-horizon closed-loop safety under disturbances.",
  "system_verdict": "partial",
  "practical_verdict": "harms",
  "mechanism_ok": 1,
  "system_judged": true
}