# Эксперимент: Criticality-Guided Failure Replay (#593)

{
  "worked": true,
  "confidence": 9,
  "verdict": "Built a finite-population criticality replay MVP with a trained auxiliary failure predictor, proposal sampling, importance weighting, ESS calculation, and replay training. Failure enrichment matched the q_alpha formula within 0.6% across alpha 0.5–4, ESS decreased sharply with alpha, and the exact importance-weighted gradient identity held to 2.3e-14 L2 error. Corrected replay matched uniform replay at alpha 0.5 and 1.0, while uncorrected replay degraded substantially; alpha 2 showed variance-related deterioration consistent with its low ESS.",
  "metrics": {
    "baseline": "Uniform replay MSE 0.165171; failure fraction 0.46058; ESS/B 1.0",
    "idea": "Critic BCE 0.35125; predicted/observed enrichment alpha=0.5: 1.399/1.402, alpha=1: 1.634/1.640, alpha=2: 1.831/1.838, alpha=4: 1.946/1.951; ESS/B: 0.696, 0.268, 0.0171, 0.0000287; corrected replay MSE alpha=0.5/1/2: 0.165123/0.165160/0.180043; uncorrected replay MSE: 0.197351/0.294067/0.385045"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 criticality_replay_experiment.py",
  "files": [
    "criticality_replay_experiment.py",
    "results.json",
    "run_output.txt"
  ],
  "limitations": "This is a controlled finite-state regression replay test rather than a full CartPole actor-critic experiment. The critic is logistic rather than an MLP, clipping and self-normalized weights were not benchmarked, and real-environment wall-clock or sample-efficiency gains were not tested."
}