How our NBA model works

1. Every final score feeds the team ratings

After each night's games, we refit ratings for all 30 teams using every final this season. The fit includes points scored and allowed, the home team and any team playing for a second night in a back-to-back schedule. Early in the season, when a few games provide little information, last season's ratings remain part of the fit. They carry 0.6 of their value, with a weight equal to 8 games. They stay in the fit until this season's results push them out.

Our fit also measures the extras. At present, playing at home adds 1.73 points, while a back-to-back game on its second night subtracts 2.08 points. As results come in, both numbers shift a little.

2. Turning a rating into a win chance and a fair line

We get the expected margin by adding the rating gap, home court and rest. NBA margins vary around that figure by about 14 points. A 4-point favourite therefore wins a bit more than 60% of the time. Both teams' offensive and defensive ratings join the league scoring average to set totals. Those totals vary by about 18.4 points.

The expected margin becomes the fair spread, while the expected score becomes the fair total. For the fair moneyline, we turn the win chance into a price that includes no book margin.

3. How we define an edge

Each price has a break-even chance. At −110, the break-even point is winning 52.4% of the time. Our chance for that side minus the price's break-even chance gives the edge. In each market, we publish and label the side with the bigger edge:

• Wide edge 6 points or more
• Clear edge 3 to 6 points
• Slim edge under 3 points

At 12 points, the bar is full. A wide edge doesn't guarantee a result. It shows where our number and the market's differ most. The market is often correct, which the picks record shows.

4. Each pick logged once, then graded for everyone to see

On game day, each pick is logged before tip with its line, price and time, and no edits are made afterward. After the final, the pick becomes a win, loss or push, graded at a flat one unit. The DraftKings closing line is compared with the line we took. We call this gap the closing-line value. If the number is positive, the market shifted toward our side after we logged it. Across many picks, that's the strongest sign a model is finding real value.

5. Season projections for the games ahead

We use the current ratings to run each remaining game 10,000 times when we work out standings and futures. From those runs we total the wins, seeds and titles, and we count the play-in spots and playoff berths. For the playoffs, we draw best-of-seven series, and the higher seed gets home court.

6. Sources for the numbers

ESPN's public scoreboard and its game summaries provide scores, schedules, team and player stats, along with rosters, venue data, injuries and the DraftKings line. Every game links to the league's official injury report PDF. Kalshi's public market data is where we get our Kalshi prices. The site doesn't yet include player prop lines. A props table appears only when we have real posted lines.