The pre-game moneyline model (Elo)

Before a ball is tipped, FinalSkore already has a view on who wins — the moneyline. That view comes from an Elo rating model, one rating per team per league. Elo is the same idea used to rank chess players: beat a strong team and your rating jumps; lose to a weak one and it drops hard.

How Elo reasons about basketball

Every team carries a number — its rating (everyone starts at 1500). To predict a game, the model compares the two teams' ratings and converts the gap into a win probability. A 100-point rating edge works out to roughly a 64% chance to win; a larger gap, a larger chance. Two adjustments make it basketball-aware:

• Home-court advantage — the home team gets a fixed rating bonus (about 100 points) before the comparison, because home teams genuinely win more often.
• Season carryover — at the start of a new season, ratings are pulled partway back toward the league average (1500), because last year's roster isn't this year's.

After each game the winner's rating goes up and the loser's goes down by a step sized to how surprising the result was: beating a much stronger team moves your rating a lot, beating a much weaker one barely at all. The step uses a fixed sensitivity (the "K-factor") and counts only the win or loss — not the final margin.

Why this captures "who beat whom"

Elo is transitive by design: if A beats B and B beats C, A's rating ends up ahead of C's without the two ever playing. It also has memory with decay — recent results move the number most, but a whole season of games shapes it. That's exactly the "recent form plus the web of results" a sharp fan reasons about, turned into one comparable number per team.

What you see on the card

The moneyline panel shows each team's Elo, the model's win probability, a confidence band, and a small decision tree — each team's recent games and the head-to-head history behind the number. Only picks in the publishable bands (65%+) are tracked. In backtests the model beat the naive "home team wins" baseline in every league and stayed stable across seasons (roughly 69% win rate at the NBA's ≥65% band, higher in college).

Elo is a rating, not a crystal ball. A 70% favourite still loses 3 times in 10 — the model is telling you the odds, not the result.