Power Rankings That Play Out the Rest of Your Season
FantasyEdge Team · Nov 4, 2026 · feature-explanation · 7 min read
Our Power Rankings no longer grade rosters as they stand. They forecast every team in your league over the nights left, with injuries and games played counted in, and rank each by its chance of beating the others.
Most power rankings answer the wrong question. They add up each roster as it stands today and tell you who has the best players. A manager in November wants something else: who is most likely to win from here, given the schedule left, the injuries already on the report and the games each player will actually play.
That is what our Power Rankings now do. Connect your ESPN, Fantrax or Sleeper league, and every team in it is forecast over every night left in the season, then ranked by its chance of beating the rest of the league: by categories, by weekly matchups, or by roto points.
Open your league's Power Rankings
Every night left is a lineup
A roster does not score; a lineup does. So the forecast plays the season the way your league does. On each night left, every team starts its best players among those whose NBA team has a game, filling your league's own starting slots, best value first. The bench and the injured list are left out, and so is a player who has no game that night. A team with two centres who play on the same nights starts one of them, and the forecast knows it.
Each game counts at the player's chance to play it. For a player on the injury report, that is our injury model's chance for that game. A player the sources have out for the season counts for nothing; one out until a return date counts for nothing before it; one listed Out with no date yet counts for nothing for a week. Everyone else is healthy, but not every healthy player plays every game, which is the next part.
How many games each player plays
Our season-games model predicts how many of his team's games left each player will play, rest and minor knocks included, and the forecast scales each player's games to that total. Under Games played you choose which number it uses. Most likely, the default, is the number of games he most likely plays to within 10%, with an overestimate counted twice as costly as an underestimate. For most players that is the healthy season, a little above the average. Expected is the model's average, which already prices in the injuries that might happen.
The spread matters as much as the number. Counting every game as its own coin flip averages a season out, and a forecast built that way is far too sure of itself. When we tested that on 2025-26, ten players' coin-flip ranges held the games they actually played in November only 28% of the time; the season-games model's own chances held them 86% of the time. Real seasons swing as a whole: a player who gets hurt misses a stretch of games in every category at once.
So the rest of the season is played out 100 times. In each one, every player's share of his team's games is drawn from the model's chances, his lineup's other players and the bench fill in where they can, and every number on the table is the average over those seasons. Under the measure the table ranks by, a small range shows where four seasons in five end.
From a week to a chance to win it
Each player's numbers come from one of five bases, chosen under Based on. Blend, the default once the season has begun, starts from his preseason projection and moves it toward what he has done so far by as much of such a start as usually lasts: minutes settle fast, free-throw percentage barely at all. You can also use the projection alone, or his season so far, last month or last two weeks.
Every Monday-to-Sunday week ahead is then a forecast of each category: what the lineup is expected to put up, and how much that could vary, from each game's noise, from the chance a player sits, and from how unsure his rate still is. A category's chance against another team turns the gap between the two into a probability, so a narrow lead earns a little more than a coin flip and a wide one nearly all of it.
A matchup is won by taking most of the categories, and categories move together: a big night of shooting lifts points, threes and field-goal percentage at once. So each team's week is simulated 1,200 times with the categories correlated the way players' box scores are, and every pair of teams is compared on the same draws.
Reading the table
Cat. win% is the share of categories a team should take off every other team in the league, a close category counting near half. Matchup win% is the share of weekly matchups it should win, and Beats counts the teams it is favoured against. Roto pts adds up its expected place in every category the way a roto table does. Rank by chooses which of the three orders the table; in a roto league it starts on roto points, scored over the rest of the season as one total.
The vs columns show every team against yours: the expected score, a square per category that is likely won or lost, and the chance of winning the matchup. Press another team's circle to see the league from its side. Below the table, Every matchup shows each team's chance against each other team in one grid.
Injury risk is the starts a bad-luck season, one in ten, costs a team's lineup against an average one, with its level against the rest of your league. Open it to see the players whose seasons swing it the most, each with his likely range of games. A team built on fragile stars can sit high on the table and still carry the most risk.
What it does not know
It assumes a daily-lineup league that starts its best players every night. It cannot see trades or pickups before they happen, and the rosters are re-read every hour from your league. A points league is still ranked on the standard nine categories. And it is new: it has not yet been scored against a finished season of real leagues, so read a forecast in November as a well-informed estimate, not a verdict.