Enhanced Fourier forecast: 4 time windows (30d/60d/90d/180d) running in parallel, adaptive harmonics based on volatility, and RSI/MA confirmation gate for calibrated confidence.
Runs FFT extrapolation on 4 separate time windows (30d/60d/90d/180d) in parallel. Each captures a different cycle scale — short-term swings vs medium vs long-term trend. Final forecast = average target + majority direction vote.
Why: v1 used only 90d. If you only look at one timeframe, you miss cycles that operate at others.
Number of Fourier harmonics adjusts to market volatility: low vol → fewer harmonics (5-7, filter noise), high vol → more (15-22, capture complexity). Each window also auto-scales its harmonic count to its data length.
Why: v1 was fixed at 10. Over-fits in calm markets, under-fits in turbulent ones.
After FFT produces forecast, confidence is dampened when: RSI > 75 (overbought reversal risk), RSI < 25 (oversold bounce risk), or when forecast contradicts 50-MA position (against-trend).
Why: v1 confidence was hardcoded. v2 calibrates by checking momentum reality vs cycle prediction.
Short windows (30d/60d) use 1h klines (more samples → better cycle detection). Long windows (90d/180d) stay on 4h klines for trend stability. Every window's forecast is shown separately — you can see exactly which agrees and which disagrees.
Why: v1 was a black box. v2 lets you debug — if 3/4 windows say UP, that's stronger than 2/4.