Good Schedule, Likely to Play: Our Games-Played Model

Every fantasy manager knows the schedule trick: pick up a player whose team plays four times this week, start him every night, and win the volume categories on games alone. It works. Until he rests the second night of a back-to-back, or sits two games with a sore ankle, and the four-game week turns into two.

A team's schedule is a fact. Whether a player is on the floor for it is a prediction. Our Good schedule signal puts the two together: it counts the games a player is likely to play from tonight on, not the games his team has, and flags him only when that number beats what a typical rotation player will play over the same nights.

Games from tonight, at his chance to play each

The signal looks at two windows: the next seven nights, and the next four. A game that has already tipped off is gone; only games still to come count. Each one is weighed by our chance he plays it, and those chances add up to his likely games.

Each chance comes from the most specific source we have for that game. For a player on the NBA's injury report, it is our injury return model's chance he plays each of his team's next five games. When the injury reports have him out that day, out for the season or before the latest return date they give, the chance is zero. For everyone else, healthy players included, it comes from our availability model, built for exactly this question.

A model for stretches, not single nights

The obvious way to estimate a week is to take a player's chance of playing one game and multiply it across the week. That gets the most important cases wrong. Rest days come with back-to-backs, and a player who sits one night of an injury often sits the next. The nights are not independent, so we built a model that predicts the stretch directly: how many of his team's next two, three, four and five games each player will play.

It learned from every game of four NBA seasons, 2021-22 through 2024-25, one row for every rotation player on every team night, healthy or on the report. Then it was tested on the whole of 2025-26, a season it never saw. Its most likely count was exactly right 77% of the time over two games and 59% over five, and its expected count was off by a third of a game over two games and under nine-tenths of a game over five, on average. That is the version running on the site.

What drives our availability models is no surprise to anyone who manages a team: a player's status on the night-before injury report, and his recent minutes and games played. Everything else they consider, from age and injury history to back-to-back habits and teammates out, moves the answer far less.

And for the weeks after

Past the next five games, a second model takes over: our season games model, which predicts the share of his team's remaining games each player will play. On 2025-26, which it never saw, its rest-of-season estimates were off by 6.7 games on average, against 8.3 for the team's own projection and more than 15 for the assumption that a healthy player plays every game. Where the stretch model has no answer for a player, his next ten games are read from it, together with the injury reports' return dates.

What counts as a good schedule

A player earns the signal when three things are true at once. His team plays at least four games in the next seven nights, or three in the next four. He is likely to play all of them but half a game. And his likely games are at least half a game more than the median rotation player's, a player averaging 20 minutes or more, over the same nights.

That last rule is what makes it a signal. In a week where most teams play four, a four-game schedule is no edge at all, and nobody is flagged. The edge is a team with more games than the league, and a player who will actually be there for them.

Both half-game bars were checked against the live predictions. A healthy rotation player is expected in about 94% of his team's games, around 3.8 of four. Half a game short keeps those players and drops the ones on the injury report and the ones likely to rest a back-to-back. A third of a game would also have dropped dozens of healthy starters. And the bars sit on a plateau: anywhere from a quarter of a game to three quarters picks nearly the same players, because a team with one more game than the league is almost a full game ahead and a team with the same number is not ahead at all.

Updated with every injury report

The availability model runs again every time the NBA publishes a new official injury report, so a player downgraded in the afternoon loses his likely games before tip-off. The season games model runs once a day. The injury reports themselves are read every hour during the season.

What no model can see

No model knows about the ankle that gets rolled in tonight's first quarter, or the coach who decides on a rest day nobody announced. Likely games are a chance, not a promise, which is why every game on our pages carries its own chance rather than a yes or a no.

Where to find it

The Good schedule badge is on our Trending Players page, where you can also order the list by most likely games, and it counts toward the day's top pickups. The home page's week ahead shows the next seven nights and the pickups likely to play the most of them. A player who carries the signal says so on his own page. And the Schedule Grid's By night view, switched to Players, shows every game ahead for each player at his chance to play it.