Your Predictions
NBA · Lineup-built lines

Set the rotation. Get the line.

Every player carries a rating for what he's worth per 48 minutes, and the minutes he actually plays. Turn players off, type in minutes, and the projected line moves with the rotation you build. It won't beat a closing line on its own, and it isn't meant to — it's here to show you what a lineup change is actually worth before you decide anything.

The matchup

Our linehome side
Fair odds
Book line—you enter it
Your edge—points of disagreement

Away rotation

Home rotation

Your call

How it could finish

chance of each final marginyour side coversbook line

Line movement

Every game we price gets logged. If the line drifts toward our side after we flag it, we saw something before the market did. That's the test running this season, and it needs no money on the table.

Power ratings

tap to load a team

Points per game each team projects to outscore an average opponent by, with the rotation currently set on this page.

Model settings

Player ratings are points per 100 possessions, built from every possession of the last three seasons with teammates and opponents held equal. A rating of +5 means his team outscores opponents by 5 more per 100 with him on the floor than with an average player there. Most rotation players sit between −3 and +4.

Home court is added to the home team's projected margin. Two points is the recent league average, and it has been shrinking for years.

Rating scale shrinks the gap between two teams before it becomes a line. Testing on four seasons found projections slightly too spread out, and 0.9 corrected it. Set it to 1 to see the raw difference.

Rookie rating is what a player with no measured possessions gets. −1.5 is what a typical first-year player is worth. They're marked in blue in the rotations.

Margin spread is how much NBA results scatter around the projection. Twelve points is the historical figure, and it's what turns a projected margin into a cover chance.

Play at gap is how far the book's line must sit from ours before the page calls it a play. Your settings are saved on this device.

How to read the gap

This number does not beat the closing line, and you shouldn't bet it as though it does. Tested on four seasons of games with every pick made from information available beforehand, it misses the final margin by 11.0 points on average. The sportsbooks miss by 10.4.

Against the spread it ran under 50% at every size of disagreement, and the gaps got worse as they got wider: 44.6% when we were 7 or more points away from the book. A disagreement that big means the market knows something we don't, not the reverse.

We also checked the obvious hope — that the model catches injuries faster. In the games where one team was missing the most minutes, it hit 46%. By tip-off the line already has the news.

Know the size of our own error. Checked against how teams actually performed last season, our team ratings miss by about 4 points per team per game. Two teams in one matchup means a gap under roughly 5 points is inside our own noise — it isn't a disagreement with the book at all, it's rounding.

What the gap is good for is triage. It tells you which game on the board is worth your attention, and the rotations tell you why. The bet itself needs something the model can't see: a rotation change you know about, a number that moved for a bad reason, a spot you understand.

Gap between our line and the bookBetsHit rate
Under 1 point1,08147.5%
1 to 3 points1,83949.0%
3 to 5 points1,12248.1%
5 to 7 points53646.6%
7 points or more28744.6%

Break-even at standard −110 odds is 52.4%. From 4,888 regular-season games, 2022-23 through 2025-26. Ratings were refit monthly using only games played before that date. An earlier version of this model, built on box-score stats and tested back to 2007-08, did show a 62% hit rate on large disagreements in the opening weeks of a season; this stronger version did not reproduce it on the four seasons it covers, so treat that as unproven rather than as an edge.

What it is and isn't

Every player carries a plus-minus rating: how many points per 100 possessions his team gains with him on the floor instead of an average player, with teammates and opponents held equal. It comes from every possession of the last three seasons — 218,000 stretches of play, rebuilt lineup by lineup.

Stack a rotation's ratings by minutes and you get the team. Subtract the two teams, add home court, and that's the line.

It picks straight-up winners 67% of the time against the market's 68%. Close, and close isn't good enough to bet blind.

What it's for is seeing the shape of a game. Bench a starter and watch the number move. Give a rookie 30 minutes and watch what it costs. That's a question the model answers well, and it's the question a betting line can't answer for you.

It doesn't do totals. It says who wins and by how much, not how fast a game gets played.