
I built a pro-draft win-probability tool based on Champion Draft for statistical enthusiasts
With MSI on I kept arguing with friends over whether a comp had "won draft," so I built a small site that estimates blue-side win probability from real pro data (Oracle's Elixir — thousands of games, patch-weighted, current patches only).
How it works:
- Pick 5 champs a side → win %, plus a breakdown of champion strength, synergy, and lane matchups.
- Optionally pick the two teams → it adds team Elo and the actual starting players' Elo + their history on the champ they're on. Rosters auto-fill from each team's last game, and you can edit a seat to sub someone in. Keyboard-friendly (type a champ, hit Tab/comma to lock it and jump to the next).
The honest part, because this sub will (rightly) ask:
- Draft alone is a weak predictor. Champions-only sits around 55% accuracy - barely above just picking the favorite - because who's playing the comp matters far more than the comp. Add the teams and it climbs to ~66%.
- It's worst at international events like MSI, where cross-region team strength is hard to calibrate (the tool flags cross-region matchups so you don't over-trust them).
- Everything's validated with walk-forward backtests (no peeking at future games), and the methodology page shows accuracy broken down by league and patch - plus a list of things I tried that didn't work.
It's a draft-analysis / what-if toy, not a betting edge - pro games are genuinely hard to predict and I'd rather be honest about that.
(Optional, caveated solo-queue line - I'd cut it, but if you want it:) A few people asked about solo queue - the draft-only champion/synergy read can loosely sanity-check your own draft, but it's trained on pro games, so treat it as a rough guide at best.
https://comp-strength.vercel.app/ (haven't bought a domain yet) - would love feedback, especially on the modeling. Hobby project.