BTC · Rolling Prony fit · Trend + complex damped-sinusoid poles
The polynomial-and-coefficients solver, refit on each anchor's own trailing 90 days (the per-candle refresh), deliberately allowed to overfit -- then tested out of sample two ways: the RECURSION (stepped forward from the last real prices, red) and the FUNCTION-OF-TIME wave (the explicit polynomial evaluated at future time points, dashed green), against flat and straight-line baselines -- plus the breakout question: at real 20-candle-breakout moments, does either variant's initial direction call beat the honest chance rate?