Forecast Model — Phase 1 Statistical Engine (EMA + Seasonal Adjustment). Algorithm: 13-week Exponential Moving Average (α=0.15) for trend smoothing → linear slope from last 26 weeks → month-of-year seasonal multiplier → 90-day projection with ±1.5σ confidence bands. Estimated directional accuracy: 70–75% based on backtesting against the last 13 weeks of price history. Phase 2 upgrade (post-Denbigh): ARIMA(2,1,1) model via Supabase Edge Function. Target accuracy: 82–88%. Same UI — data source switches transparently.