Эксперимент: Koopman Preview Gate for Adaptive Neural Computation (#823)

{ "worked": false, "confidence": 9, "verdict": "Built a reproducible NumPy Koopman preview gate with quadratic risk, threshold sweeps, bounded-rate interpolation, and baseline/always-refine comparisons. Exact spectral-decay and analytical preview-lead predictions were confirmed, and gating saved 83.5% of refinement calls with 98.8% recall; however, gated MSE was 10.7% above always-refine, exceeding the requested <5% degradation, and the learned autonomous model did not show empirical lead time on the burst-driven system.", "metrics": { "baseline": "MSE 0.0013426; refinement calls 100%", "idea": "MSE 0.00001742; 16.5% refinement calls; 83.5% calls saved; 98.8% recall; 10.7% increase vs always-refine MSE", "always_refine": "MSE 0.00001573; 100% refinement calls", "spectral_decay": "For radii 0.50, 0.80, 0.95, observed log slopes matched log(r) to <6e-16 relative error", "analytic_preview_lead": "For growth factors 1.02, 1.05, 1.10 and horizons 2, 5, 8, observed lead exactly matched predicted H", "threshold_sweep": "At 82nd-percentile threshold: 17.95% gate rate, 99.6% recall, precision 1.0" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 koopman_preview_experiment.py", "files": [ "koopman_preview_experiment.py", "results.json", "run_output.txt" ], "limitations": "This is a small deterministic NumPy toy experiment, not a trained GRU/encoder or language-model benchmark. The encoder is identity, Q is fixed to the identity, no learned differentiable gate or scheduling variable was tested, and the burst forcing is not included in the autonomous Koopman preview model; CUDA and wall-clock compute savings were not measured." }