tennis projections
resolve tennis
ATP + WTA tournament Monte Carlo from our own surface Elo + Bayesian aging-curve posterior + statistically significant archetype matchup matrix. Currently shipping the 2026 French Open reconstruction (Zverev ranked #2 P(WIN), he won).
French Open 2026 — ATP reconstruction
Clay · 128-draw bracket Monte Carlo · 5,000 sims · top-15 win odds. Sinner J 59% / Zverev A 8% / Ruud C 6%. The model ranked Zverev #2 by P(WIN) entering the tournament he actually won.
open
how it's built
- Surface Elo — own build: K=32 × tier × margin scaling on Sackmann ATP 1968-2024 + tennis-data 2025-2026, 3 ratings/player (hard/clay/grass)
- M1 aging posterior — pymc hierarchical (NBA pattern): per-surface quadratic age curve + 1208 player-level age tilt. +0.92pp acc OOS on ATP
- K=7 archetype matchup — physical (height, lefty) + box-score serve outputs + MCP behavioral (volley%, net rate). Net-rusher splits into Big-Server (Sampras/Isner/Karlovic) vs Classic S&V (Becker/Henman/Rusedski)
- 4 Bonferroni+FDR-passing edges: Big-Server > Counter-Puncher on CLAY (+5.3pp, p=0.00024) and Modern Baseliner > Crafty Veteran on GRASS (+6.5pp, p=0.00107)
- Walk-forward OOS validated through 2025 + 2026-YTD: 63.97% ATP acc / 0.6276 LogLoss / 0.2194 Brier