The Volatility Channel (Financial Branch)
🔗 Reproducible code: github.com/jpbronsard/syntonic-portfolio v 3.0 V2.0 measured one channel of financial adaptation: the return channel, where \(\tau^\star=1/\sqrt{2}\) is the structural signature of the random-walk limit. Markets have a second channel with the same structure as deep learning's variance stream: realized volatility, estimated from squared returns, is a multiplicative (\(\chi^2\)-class) process, and the industry has run a hand-tuned window on it since 1994, the RiskMetrics EWMA \(\lambda=0.94\), about 16.7 trading days. This paper applies to that channel the same conforming estimator (log domain, scale-matched anchors, TH43/H16 noise class) that recovered Adam's variance window in the deep-learning validation. The answer comes in two parts. Under the diffusive read, the industry window is not recovered: the measured window is 48.5 days on equities and 55.3 on gold, outside the pre-registered [10, 40] band. The verdict is GREY and is reported as-is; the instrument is healthy, with trigamma stability 0.11 to 0.14. The miss, however, lands where Theorem U' points. The roughness measurement, whose band was frozen before the run, returns \(\widehat H = 0.169/0.149/0.097\) across the three assets, all inside [0.05, 0.20]: the volatility channel is off-manifold, the diffusive read is inflated accordingly, and this is the first real-data instantiation of U''s documented exception class. The quantitative reconciliation, the fractional horizon formula, is stated as the named open problem. Three modules sharpen the result. The \(\tau^\star(s)\) spectrum converts the degenerate scalar into a falsifiable curve and finds it flat at \(1/\sqrt{2}\) at all eight scales from 1 to 252 days, every confidence interval containing the floor. The C41 coherence gain of V2.0 is closed against TH41-B: Monte Carlo at the measured tail ratio \(\rho=1.320\) gives a 9.6% estimator-variance reduction bracketing the observed 1.2 to 2.1% gains. And the analytic layer, the spectrum identity, its IID degeneracy, the RiskMetrics window bounds, and the per-asset U' exponents, is machine-checked in Lean 4 with Mathlib, no sorry, standard axioms only; the formalization also shows the constraint \(\rho_1<1\) is unnecessary. All verdict bands were pre-registered and hash-sealed before any market data was touched. No trading claim is made: every result is an estimation statement. Version 3.0, companion to V2.x. Concept DOI: 10.5281/zenodo.18642832. RETURN TO HUB : https://doi.org/10.5281/zenodo.17254395