# Эксперимент: PAC transition-cover training monitor (#835)

{
  "worked": true,
  "confidence": 8,
  "verdict": "Built a PAC transition-cover monitor with deficit-first cell allocation, reproducible Bernoulli miss-probability verification, and an equal-budget MLP transition experiment. The math mechanism manifested: across 20 (epsilon,n) settings, empirical miss probabilities matched (1-epsilon)^n with maximum absolute error 0.00444, and all four PAC threshold tests were below delta=0.1. The neural comparison showed a modest improvement for PAC collection (test MSE 0.01174 vs 0.01194; cell-3 MSE 0.07965 vs 0.08897), but this is only a weak toy signal rather than a demonstrated general sample-efficiency win.",
  "metrics": {
    "baseline": "Uniform collection, 640 transitions, test MSE=0.0119438, cell-3 MSE=0.0889726, rare-branch MSE=0.6314883, rare samples=9",
    "idea": "PAC deficit-first collection, 640 transitions, test MSE=0.0117403, cell-3 MSE=0.0796502, rare-branch MSE=0.5347307, rare samples=10; required n=24 per cell and all cells covered"
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 pac_transition_cover.py",
  "files": [
    "pac_transition_cover.py",
    "results.json",
    "run_output.txt"
  ],
  "limitations": "Only a scalar synthetic transition system and one fixed seed were tested. The prototype uses total sample count as the PAC readiness condition and does not implement a learned outer successor envelope, conformal calibration, held-out violation monitoring, dependent-sampling corrections, long-horizon rollout evaluation, multiple seeds, or a realistic MPC/world-model workload."
}