← Research

Is loser’s queue real?

Published September 23, 2026 · NA ranked solo queue, 563,839 games and 5,638,390 player-games, September 2 to 9, 2026

What we measured

Loser’s queue is the belief that after a winning streak the matchmaker hands you weaker teammates to drag you back to 50%. If it were real, it would leave a mark in the games: players on a win streak would lose their next game more often than their form predicts, and their teammates would rate below their opponents.

• We built a model that sees only what exists when the queue pops: the ten players’ ranks, their form over the last 7 days and the current session, autofill and duos. It sees no champions and no draft.
• We trained it on about 5.1M NA ranked games from June to August 2026 and tested it on a week it had never seen: September 2 to 9, 563,839 games and 5,638,390 player-games (one per player per game).
• For every player-game we recorded the streak the player walked in with, whether their team won, what the model expected, and how their four teammates rated against the five opponents, not counting the player.

Streaks carry on. They don’t flip.

If there were a loser’s queue, the win-streak groups would fall below 50%. They rise instead: from 50.4% after one win to 53.0% after five or more. On the losing side the rate barely moves until five losses in a row, where it drops to 48.4%.

The simplest explanation: a player on a win streak is often better than their current rank and keeps winning until the rank catches up, and a player on a long losing streak is, for now, playing worse.

The model saw it coming

In every streak group the actual result lands within half a point of the model’s prediction. The model knows nothing about any punishment. It only knows the players’ recent form and ranks, and that is enough to explain the outcomes. No hidden mechanism is needed.

Your teammates don’t get worse after a win streak

We compared the average rating of a player’s four teammates with the five opponents’, leaving the player out. After five or more wins the teammates rate slightly above the opponents (+0.126 on our rating scale); after five or more losses, slightly below (-0.061). Both gaps are tiny, and neither goes the way loser’s queue claims.

So what does put you on the weaker team?

For every player-game we measured two things: how often the player’s team started as the underdog (under 50% before champion select), and how much of that came from the matchmaker: the teammates’ average rating against the opponents’, not counting the player.

• The biggest factor is you. After a losing streak your team is the underdog more often because the model rates you lower: your last few games went badly. Your teammates aren’t worse. The matchmaker’s part barely moves, and it moves the other way.
• Autofill costs about a point of win rate (49.1% against 50.1%), and nothing compensates for it.
• The best player in the lobby gets slightly weaker teammates. The matchmaker balances team averages, so the strongest player’s four teammates rate a little lower. The feeling of having to carry every game is real. It’s just small.
• There’s no promo queue. One win from promotion or one loss from demotion, the odds are the same as anywhere else in the division.
• Nothing else moves it: not your LP trend, not how deep you are into a session, not how fast you requeue.

Player by player

We looked at the 194,935 players with 10 or more games that week. The share of games in which a player’s team started as the underdog ranged from 31.6% to 68.4% (10th to 90th percentile): more spread than luck alone produces, and it mostly follows the player’s own form. A player’s actual win rate in the first half of their week predicted the second half almost not at all. Over one week, luck dominates.

Other things we found

• About one ranked game in five (21.2%) has a real favourite at 60% or more before anyone picks a champion. Those favourites win about as often as predicted: the ones given 70% or more won 76.4% of the time.
• Low elo is much less even. More than half of Iron lobbies have a favourite at 60% or more, against 9.4% in Diamond.
• Autofill isn’t balanced between the teams either. When blue has two more autofilled players than red, blue wins 47.1%; the other way round, 52.5%.
• Blue side usually gets the lower-rated team (lower average LP in 61.3% of games) and still wins about half the time (50.1%).

How good is the model?

The players-only model beside three simpler forecasts, on the same 563,839 test games, in the five measures we report for every model (the accuracy page explains each).

Our own rating gap alone predicts almost nothing (AUC 0.5023): by that rating, the matchmaker balances the two teams nearly perfectly. What the model adds is recent form; the last 7 days carry about half of what it knows.

What this can’t tell you

• NA only, one week, one patch. Streaks are counted within the games we collected for each player, so a game we missed can split a streak in two.
• “Rating” is our own estimate, built from match history. It is not Riot’s hidden MMR, and this is not a look at the matchmaker’s code: it measures lobbies as they arrived, using public information.
• These are averages over millions of games. Your own bad night can still be a bad night.