Эксперимент: Lipschitz Forward-Invariant Policy Certification (#947)

{ "worked": true, "confidence": 9, "verdict": "Built a scalar closed-loop Lipschitz forward-invariance certification MVP and verified its core mechanism quantitatively. The perturbation bound matched sampled maxima within 0.8%, the predicted disturbance boundary matched the observed grid boundary within 0.44%, and certified-margin loss scaled exactly linearly with policy gain. The baseline and certified training runs were identical because the certificate penalty was inactive, so this validates the math but does not establish an RL performance win.", "metrics": { "baseline": "Lipschitz bound 1.5366; certified margin 0.7185; mean nominal violation 0.01135; action sensitivity 0.5853", "idea": "Lipschitz bound 1.5366; certified margin 0.7185; mean nominal violation 0.01135; action sensitivity 0.5853. Bound/sample ratios were 0.9926-0.9998, boundary relative error was 0-0.44%, and gain scaling matched the predicted linear drop exactly." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 certify_experiment.py", "files": [ "certify_experiment.py", "results.json" ], "limitations": "This is a scalar toy system rather than PPO/SAC or the proposed HVAC task. Spectral normalization was measured but not enforced, the certificate penalty remained inactive, and no multi-step stochastic RL, learned dynamics, nonlinear safety constraints, or GPU experiment was tested." }