The 2026-27 NBA Season, Simulated Thousands of Times: Projected Standings and Playoff Probabilities

October 19, 2026

Opening night is Tuesday. Before a single possession is played, here is what a Monte Carlo engine — one with a published accuracy record on last season — makes of the 2026-27 schedule: the 1,200 games that already have opponents, simulated one by one, with projected wins for every team, the probability each one reaches the playoffs, and the shape of both conferences.

The method is the same one behind every DunkSim projection, and the limits are the same too. Both are laid out below, because a standings table without its error bars is just a prediction, and this isn’t meant to be one.

How this was built

1. Every scheduled regular-season game — the 1,200 that have opponents today — was simulated with the rosters as they stood on the evening of October 19, 2026 (5:19 PM Central), after the final cut-downs. Each team’s rotation was trimmed to its ten highest-minute players before simulating, a ten-man rotation being roughly the shape of a real game night. The ten were ranked on last regular season’s per-game minutes, never on preseason minutes; a player with no NBA minutes yet was slotted by his draft position, which puts at most one rookie into a team’s ten. Each game was played possession by possession ten thousand times, the same count as the accuracy test, and the share of runs the home team won is that game’s win probability. (How the simulation works.)
2. Injuries were applied by date, not frozen. A player listed out with a reported return date sits out the games through that date and plays after it. A player out with no return date, with a season-ending designation, or with a return date after the regular season ends, sits out all 82 — those players are listed below, because they are the absences that move a win total. Anyone day-to-day, questionable, probable or doubtful was treated as available for every game. Rest days and load management are not modeled. Sources: the Official Injury Report and reported return dates as of the evening of October 19.
3. A team’s projected wins is the sum of its win probabilities across its games. A team at 60% in every game projects to about 49 wins; the real number scatters around that.
4. Playoff and play-in probabilities come from sampling whole seasons: draw a result for every game from its win probability, build the standings, apply the play-in format, repeat ten thousand times. Ties are broken by head-to-head record among the tied teams in that sampled season where the teams have at least one scheduled meeting, otherwise at random. Each play-in game is drawn from the engine’s own probability for that pair, with the higher seed at home. “Playoff probability” below means finishes in the top six of the conference, or wins through the play-in. “Play-in probability” means finishes seventh through tenth.
5. Thirty games aren’t scheduled yet. The NBA Cup’s knockout rounds and the make-up games the league assigns after group play (two per team for the 22 teams that don’t advance, one for each quarterfinal loser) won’t have opponents until late November. For this run each team’s two unscheduled games are counted at a 50% win probability — one win each — so the table reads as 82 games. The most that can move any team is a fraction of a win. The re-runs will simulate those games with their real opponents once the league publishes them.
6. No adjustments. These are the engine’s baseline reads. Nobody put a thumb on any team.

Treated as out for the season in this run: Henri Veesaar (Hawks), Otega Oweh (Thunder), Ariel Hukporti (76ers) and Ben Simmons (Kings). None carries a season-ending designation; each has a reported return date that falls after the last regular-season game, so he misses all 82 in the projection.

Treated as out through a reported date (the ones a reader is likely to ask about; 31 in all): Jimmy Butler III (Warriors, through January 2), Kristaps Porzingis and Moses Moody (Warriors, both December 1), Bradley Beal, Brandon Ingram and Max Strus (Clippers, all November 8), Donte DiVincenzo (Timberwolves, April 2), Shaedon Sharpe (Trail Blazers, March 2), Mark Williams (Suns, February 3), Jusuf Nurkic (Jazz, November 4), Nic Claxton (Bulls, November 4), Isaiah Stewart (Grizzlies, October 23), Julian Strawther (Nuggets, November 29).

Rosters change. This table is a dated snapshot, and the date is in the header. We’ll re-run it at the All-Star break and before the trade deadline and keep the earlier versions on this page.

Eastern Conference

Western Conference

Projected W–L is rounded and includes two unscheduled NBA Cup-related games per team counted at 50%; playoff, play-in and lottery percentages are rounded and sum to 100 for each team. Snapshot: rosters and injury designations as of October 19, 2026, 5:19 PM Central.

Three things the table says

The gap at the top. The West has two teams the engine cannot see losing the playoffs: the Lakers at 61 projected wins and Denver at 60, both at 100%, with a seven-win drop from Denver to Portland behind them. Portland and Utah still sit above 99%, and Portland does it with Shaedon Sharpe listed out until March, so that number rests on the roster around him. In the East, Cleveland leads at 56 wins, five clear of Indiana, and those two are the only East teams above 95%.

The crowded middle. This is the finding that matters most for reading the table honestly. In the East, seeds 6 through 9 (the Knicks, Hawks, 76ers and Raptors) are separated by about two projected wins, with playoff probabilities from 83% down to 66%. In the West, seeds 6 through 8 (Golden State, San Antonio, Memphis) sit within a single win of each other, and Minnesota at 49% is the coin flip. Eight teams across the league sit between 25% and 75%. That is the engine’s way of saying the play-in will decide a quarter of the postseason field and it does not know how.

Where it differs from last season. The rosters moved a lot this summer, and the table moves with them: projected wins correlate only weakly with last season’s records, at 0.33 across the thirty teams. The clearest drops are Oklahoma City, 64-18 last season and 38 projected wins with a 27% playoff probability now, without Dort, Wiggins and Isaiah Joe, with a rotation of nine returning players and one rookie; Boston, 56-26 and now 35 wins and 12%, without Jaylen Brown; and Houston, 52-30 and now 37 wins with essentially the same roster, where Fred VanVleet is back from a season missed entirely and the engine’s read of the veterans around Durant is lower than last season’s record implies. The clearest climbs are Utah, 22-60 and now 51 wins and 99%, with Jaren Jackson Jr. beside Markkanen; Indiana, 19-63 and now 52 wins and 99%, with Haliburton back; and Dallas, 26-56 and now 47 wins. Read every one of those for what it is: the engine’s read of each roster as it stands tonight, not a forecast of a collapse or a leap. Nothing in the injury inputs drives the big drops (Oklahoma City’s only long absence is Otega Oweh for the season, Boston’s is Devin Carter until November 8), so if you think the engine has a team wrong, the case to make is which of its games are wrong, and the simulator is built for exactly that. Golden State is the counter-case: 43 wins and 71% with Jimmy Butler out until January 2 and Kristaps Porzingis and Moses Moody until December 1 already in the run; if you think Butler returns sooner, add a win or two in your head, and if you think he doesn’t return at all, take several off.

How much to trust this

The engine’s record on real games is public and it is worth reading before you argue with any row above. On a full replay of the 2025-26 regular season, it named the correct winner in 69.3% of 1,230 games. Choosing the team with the better record on game day would have scored 67.0% on the same games. That is the honest size of what a possession-level model adds over “the better team usually wins”: real, and modest.

That test used the lineups that actually took the floor. A season projection has to guess rotations months out, so expect it to do worse than the game-level number, which is why the end-of-season scorecard matters.

A season table can also be tested as a season table. We projected last season the same way, from its opening-week rotations, and compared it with the final standings: projected wins correlated 0.56 with the real win totals and missed by about 10 wins per team. Simply carrying every team’s previous record forward would have scored 0.50 and missed by 10.6. So the method beats the lazy baseline, and a ten-win miss on a typical team is the honest size of its error.

Three things follow for a season table:

The middle is noise. A game-level model that is right seven times in ten produces season win counts that scatter several games either side of the projection. Two teams three projected wins apart are not meaningfully separated. Read the playoff percentages, not the rank order.

The top is probably overstated. The accuracy page shows the engine is overconfident in its surest calls and that its projected margins run wider than real ones. Strong teams in this table are likely a little less strong than shown, and 95% playoff probabilities should be read as “very likely,” not “done.”

It only knows what it was told on the evening of October 19. A trade, a long injury, a rookie who turns out to be real — none of it is in here until the re-run. The table is a baseline for the argument, not the end of it.

Run the parts you disagree with

Every row above is the sum of game simulations you can run yourself. Open any matchup, sit a player out, change a rotation, and see the win probability move — that is what the simulator is for, and Roster Lab lets you re-run a team with a trade you think is coming. If you believe the engine is wrong about your team, the quickest way to make the case is to show which games it has wrong.

Updates: All-Star break re-run · trade-deadline re-run · end-of-season scorecard, where this table is graded against the real standings the same way the engine is graded on the accuracy page.