How Ranked Elo Works

The live formula, in full

Ranked currently covers one queue: Gunfight 2v2. After every ranked match the server runs classic Elo on team averages — your rating and your teammate’s are averaged, the enemy pair likewise, and the gap between those two numbers sets your expected win chance. Every point one side gains, the other side loses. The complete rule set:

• Expected win chance = 1 / (1 + 10^((enemyAvg − yourAvg) / 400)). Equal teams → 50%. A team 200 points heavier → ~76%. 400 heavier → ~91%.
• Rating change = round(32 × (result − expected)), where result is 1 for a win and 0 for a loss. K = 32 is the most either team can swing on one match (before the new-player multiplier). Both players on a team get the same change.
• Wins pay at least +6. The player-voted win floor (July 2026): if the raw math gives a heavily favored winner less than +6, it is raised to +6 before the new-player doubling. Losses are never floored.
• Draws move nobody. Textbook Elo would pay the underdog on a draw; we zero it instead because “we tied, why did I lose rating?” felt wrong.
• New players ramp fast: during your first 5 ranked matches every change is doubled, so you reach your real level sooner.
• Everyone starts at 1000. Ratings are clamped between 0 and 9999, and each season gets its own rating row.
• Leavers don’t profit: leave a match your team goes on to win and your gain is zeroed. Leave a losing match and you still take the full loss (plus the usual abandon penalties). And since July 2026 the player you abandoned gets leaver forgiveness: their loss is halved when a solo-queue teammate bails early — premade partners don’t qualify and it’s capped at twice a day, so it can’t be farmed.
• Only the outcome counts. Kills, damage, and round margin don’t change the math (yet) — that keeps the numbers predictable before the match even starts.

That last property is also the honest explanation of the “−30 then +3” story: Elo pays out surprise. If the system already expects you to win 90% of the time, winning confirms the prediction (tiny gain) while losing shocks it (big loss). The two numbers used to add up to exactly K across a win/loss pair against the same lineup; the +6 floor now tilts lopsided matchups slightly in the winner’s favor. Whether that trade-off feels right at our ladder’s size is exactly what we want feedback on.

The party question

Today the formula treats a premade duo and two solo players identically — party status is not part of the math. But premades communicate, play set strategies, and consistently outperform what their raw ratings predict. If that edge is real at our ladder’s size, premades are slightly over-rewarded and solo players slightly over-punished.

One candidate fix is the party handicap in the simulator above: when computing win chance, count a premade team as +N Elo stronger than its real average. The premade then earns a little less per win and risks a little more per loss, while solos facing them get the underdog’s side of the bet. Two premades cancel out. The open questions: what should N be, should it scale with rating, and is the premade edge even measurable in our match data? We’re mining match history for the actual premade-vs-solo win-rate gap before committing to a number.

Ideas on the table

• Win floor — shipped July 2026. The community voted +6, and every win now pays at least that. First idea from this page to make it into the game.
• Party handicap — premades counted as +N Elo stronger (see above).
• Teammate-leaver forgiveness — shipped July 2026. The abandoned teammate’s loss is now halved (solo queue only, with early-leave and twice-a-day gates so queue partners can’t farm it by taking turns).
• Per-player expectation — compute your change from your own rating instead of the team average, so a high-rated player carries more risk than their lower-rated partner.
• Margin of victory — closer matches move fewer points. Powerful but easy to distort; parked for now.