Methodology

π̂ = C ∘ M ∘ Φ // Φ : 𝔻 → ℝⁿ M : ℝⁿ → Δ³ C : Δ³ → Δ³

Data flows left to right. Six inputs enter the feature stage, get combined by an ensemble of three models, then calibrated before being emitted as match-day predictions (WDL, OU, EV).

ρ̂_ij(x_m) // re-estimated per match · per-match values private

Each cell shows how strongly two features move together for a given match-up. The matrix is recomputed every match — feature weight is context-dependent.