Methodology

Win projections

Win projections estimate each team's final regular-season record by combining their current win-loss record with the expected outcomes of all remaining games. Rather than simulating thousands of seasons, we use a direct fractional approach: for each unplayed game, we compute the probability that the home team wins, then add that fractional win to the home team and the complementary fraction to the away team.

The result is a projected win total to one decimal place (e.g. 63.2-18.8). This is statistically honest since every game produces exactly 1.0 total wins between the two teams and league-wide projected totals always balance. Due to the fact we round these values to one decimal place in the UI, there can be small rounding errors and we won't always arrive at exactly 1,230 regular season games (eg. 51.04 wins gets rounded down to 51.0).

How game probabilities are computed

For each remaining game, we compute the win probability using the same Elo-based model described on the win probability page. The key inputs are each team's current Elo rating and any applicable rest adjustments for back-to-back games.

The Elo difference between the home and away team is fed into the win projection model which outputs a probability between 0 and 1. Home court advantage is built into the Elo system itself, so it's automatically reflected in these probabilities. Projections use pure Elo rather than the Elo/odds blend used on individual game win probability percentages because betting odds are only available close to tip-off and can't cover an entire remaining schedule.

Back-to-back rest adjustments

The projection engine accounts for back-to-back games in the remaining schedule. Using the same rest penalties described on the win probability page, each game's probability is adjusted based on whether either team is playing on consecutive days and whether travel is involved. This means teams with more remaining back-to-backs will see a small penalty reflected in their projected win total.

Pre-season market blending

At the start of each season, Elo ratings are initialized by blending the prior season's Elo with a market-implied Elo which is derived from pre-season betting over/under win totals. This addresses the gap between our purely mechanical Elo regression and the market's awareness of off-season roster changes.

For the first 30 games of a season, the projection engine also applies an early-season fade that blends the team's current Elo with the market-implied Elo. This prevents projections from being overly reactive to small sample sizes early in the season. By game 30, actual results fully dominate and the pre-season signal fades entirely.

Strength of remaining schedule

Alongside projections, we compute a strength of remaining schedule rating for each team. This is the average Elo rating of all remaining opponents, ranked from hardest (1st) to easiest (30th).

We use Elo rather than opponent win percentage because Elo is a better measure of current team strength. It captures recent form, accounts for strength of opponents transitively, and reflects a team's true level better than their record alone. A surging team that started slowly will have a higher Elo than their record suggests, which is the correct signal for schedule difficulty.

What projections don't account for

Win projections are based on each team's current Elo rating, which reflects their performance to date. They do not account for future injuries, trades, roster changes, or load management decisions. For near-term games where betting odds are available, a game's win probability incorporates those odds to capture situational factors. The season-long projections, however, use pure Elo because betting odds are only available close to game time.

Up next: Leverage