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

(универсальный)

Ответ агента:

{
  "worked": false,
  "confidence": 9,
  "verdict": "Implemented certified dual-price routing on the structurally matched router_regime_regression neural MoE benchmark. The best idea setting was significantly worse than the tuned baseline, and the trained-model mechanism signature did not confirm the predicted price/load relationship.",
  "metrics": {
    "baseline": "Best lr=0.01, balance_weight=0.0; mean test MSE=0.587452",
    "idea": "Best lr=0.01, rho=0.001; mean test MSE=0.664206",
    "comparison": "Paired delta=+0.076754, 0/8 wins, permutation p=0.0081; idea worse (significant)"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_bench.py",
  "files": [
    "stage2_bench.py",
    "bench_report.json",
    "run2.txt"
  ],
  "limitations": "The experiment used a small synthetic four-expert regression benchmark rather than a full Transformer language-model MoE. Top-2 routing, EMA smoothing, wall-clock latency, FLOPs, and hardware buckets were not tested.",
  "bench_report": {
    "bench_version": 1,
    "track": "router_regime_regression",
    "model": "custom_routed_experts",
    "metric_direction": "lower is better",
    "n_seeds": 8,
    "baseline": {
      "best_cfg": {
        "lr": 0.01,
        "balance_weight": 0.0
      },
      "mean": 0.5874520391225815,
      "sweep_lrs": [
        0.001,
        0.003,
        0.01
      ],
      "sweep_balance_weights": [
        0.0,
        0.01,
        0.1
      ]
    },
    "idea": {
      "best_cfg": {
        "lr": 0.01,
        "rho": 0.001
      },
      "mean": 0.6642061397433281,
      "rho_sweep": [
        0.001,
        0.01,
        0.1
      ]
    },
    "comparison": {
      "delta_mean": 0.07675410062074661,
      "idea_wins": 0,
      "n_pairs": 8,
      "p_value": 0.0081,
      "verdict": "idea worse (significant)",
      "system_worked": false
    },
    "mechanism_signature": {
      "confirmed": false,
      "mean_price_excess_correlation": -0.65,
      "dual_gaps_large": true
    },
    "custom_track": {
      "name": "router_regime_regression",
      "file": "bench/custom_tracks/router_regime_regression.py",
      "domain": "moe-routing"
    }
  },
  "system_verdict": "failed",
  "practical_verdict": "harms",
  "mechanism_ok": 0,
  "system_judged": true
}