# Эксперимент: Tempered-Stable Volatility Clock for Sequence Diffusion (#453)

{
  "worked": true,
  "confidence": 9,
  "verdict": "Built a tempered-stable AR(1) volatility-clock sampler, forward-noise constructor, analytic verification harness, and blind toy denoiser comparison. The Laplace transform, stationary mean and variance, excess kurtosis, and geometric squared-noise autocorrelation matched theory across parameter sweeps. The denoiser showed a small favorable signal, but this is not evidence of a decisive modeling win.",
  "metrics": {
    "baseline": "IID Gaussian: K=-0.0085, lag-1 squared-noise ACF=-0.0013, toy held-out MSE=0.5250 +/- 0.0025.",
    "idea": "Clock: observed K=0.7783 versus predicted 0.7721, lag-1 ACF=0.0698 versus predicted 0.0650, toy held-out MSE=0.5176 +/- 0.0042."
  },
  "how_to_run": "/home/maxwelhelp/main/bin/python3 volatility_clock.py && /home/maxwelhelp/main/bin/python3 denoiser_compare.py",
  "files": [
    "volatility_clock.py",
    "denoiser_compare.py",
    "results.json",
    "denoiser_results.json",
    "REPORT.txt"
  ],
  "limitations": "The denoiser test is a small synthetic blind noise-prediction task, not a full diffusion reverse-sampling, likelihood, financial-data, or multi-seed evaluation. Efficiency under difficult tempered-stable parameters and GPU acceleration were not tested."
}