Tentang model

Skor tenis langsung, peluang menang & prediksi — tanpa iklan.

Prediction model

Every player carries four Elo ratings — overall plus hard, clay and grass. Pre-match win probability uses a surface-blended rating (½ overall + ½ surface) in the classic Elo formula:

P(A beats B) = 1 / (1 + 10^((Elo_B − Elo_A) / 400))

Ratings update after every match with a decaying K-factor 250/(n+5)^0.4, so newcomers move fast and veterans stay stable. On recent tour matches the model is correct about 64.5% of the time (Brier ≈ 0.22) — better than seedings, honestly short of betting markets.

Head-to-head & market

For each fixture the Elo number is blended with the players' prior head-to-head record (its weight grows with the number of meetings, up to ~50%). Where odds are available we also show the de-vigged betting-market implied probability — the market is the toughest benchmark, so comparing the three (Elo, H2H, market) is the honest way to read a match.

Live win probability

During a match we run a hierarchical Markov chain (point → game → set → match), fed the live score and who is serving, calibrated so it starts exactly at the pre-match number and updates as the match unfolds. Point outcomes aren't truly independent, so treat it as a model estimate, not a guarantee.

Play

Predict the winner of any real upcoming or live match. When it finishes you score 10 points for a correct call, plus an upset bonus when you back a player the model rated below 50%. Compete against other players and 65 AI personalities on the global leaderboard.

Data & ethics

Live scores come from ESPN's public feed; results, rankings and surface stats from free public data (tennis-data.co.uk); Elo is computed by us. No ads, no trackers, no betting funnel.