A simple formula, borrowed from chess in the 1960s, lets matchmaking systems turn each match result into a new estimate of a player’s true skill. Arpad Elo, a physics professor and U.S. chess master, proposed the system in 1960, and FIDE made it the international standard a decade later. Today it powers matchmaking in League of Legends, Valorant, Rocket League, Counter-Strike 2, and Chess.com.
If you’ve ever queued into ranked and wondered what your number really means, this guide explains the origins, math, and matchmaking quirks behind Elo ratings across today’s top competitive games.
The Origin of Elo and Why It Outlasted Its Sport
Arpad Elo was a physics professor and a strong U.S. chess master in the late 1950s, and the existing U.S. Chess Federation rating system leaned on subjective votes from organizers. Tournament directors guessed who belonged where, and top players sat idle while lower-rated opponents kept matching up. Elo proposed a math model in 1960 that turned each match result into a predictable rating change for both players, and the federation adopted it that same year.
The system crossed the Atlantic in 1970 when FIDE made it the official international standard, and Elo published The Rating of Chess Players, Past and Present in 1978 to lock down the math. The core premise was practical: rating should predict outcomes, and each match should move both numbers toward a better estimate of real skill. That predictive property is why the formula still runs matchmaking engines more than sixty years later.
From chess clubs to online servers
That rating-as-forecast idea is what made the system portable. Online chess servers needed to pair strangers fairly across continents, and Elo offered a working model by 1986. Once game developers saw how cleanly it matched opponents, porting it to other competitive formats took only small tweaks, because the math already solved ranking for players who had never met.
- 1960: Arpad Elo proposes a math-based rating system to the U.S. Chess Federation.
- 1960: The U.S. Chess Federation formally adopts the Elo system for domestic events.
- 1970: FIDE adopts Elo as the international standard for world-level play.
- 1978: Elo publishes The Rating of Chess Players, Past and Present, cementing the math.
- 1990s–2000s: Online chess servers and early competitive games adopt Elo-derived matchmaking.
- 2010s–present: Major esports titles run modified Elo engines under cosmetic rank tiers.
The Math Behind Every Win and Loss
Expected score is drawn from the rating gap on a logistic curve, so a 200-point favorite wins about 75% of the time, a 400-point favorite around 91%, and a 600-point favorite north of 95%. After each match, the system compares the actual result to that expectation, and the difference drives the update. A higher-rated player who wins gains only a few points, while a higher-rated loser surrenders a much larger chunk, and the change is zero-sum so the population average stays stable.
Skill sits on a logistic-like scale, not a percentage, which is why a 100-point gap means the same thing at 800 or 2800. That property keeps the formula portable across skill tiers, so the same math works for a Bronze placement match and a grandmaster tournament. A 1500-rated player isn’t “twice as good” as a 750-rated player; rating simply maps to win probability, and a 200-point gap still predicts about three wins in four.
What the numbers mean in practice
That scale is what makes Elo behave the same way from a casual online match to a titled tournament. A 100-point gap is a 64% favorite whether the players sit at 900 or 2900, which is why developers can reuse the formula without rewriting it per tier. The practical takeaway: rating gap, not raw number, is the unit the system actually thinks in.
| Rating Gap | Expected Win % (Higher-Rated Player) | Expected Loss % (Lower-Rated Player) |
|---|---|---|
| 0 | 50% | 50% |
| 100 | 64% | 36% |
| 200 | 75% | 25% |
| 300 | 85% | 15% |
| 400 | 91% | 9% |
| 500 | 95% | 5% |
What the K-Factor Actually Controls
The K-factor is the multiplier that caps how many rating points a single match can move, and it’s the main dial developers tune per game. A high K-factor makes ratings swing fast after every match, while a low K-factor means a single upset barely dents a career number. K-factor is the simplest tool for shaping how quickly the system finds your real level.
Most competitive games run two or three K-factor tiers at once. New or provisional players usually sit on a high K-factor, often 30 to 40, so the system finds true skill within ten to fifteen games. Established players settle into a lower K-factor, commonly 10 to 20, which prevents a lucky weekend from inflating a rank or a bad night from destroying one. Some titles also adjust K-factor dynamically based on rating tier, games played, or the system’s confidence in your skill.
Why a K-factor shift feels personal
The moment your account crosses from provisional to established, the same win or loss starts moving a smaller number. That shift is the K-factor dropping, which is why a five-game win streak feels huge in placement matches and modest a month later. The system is running the same math; it’s just protecting the established rating from noise, the way a car’s suspension absorbs a speed bump.
Heads up: a sudden K-factor reduction is the most common reason your gains “slow down” overnight. The math didn’t change, only the cap.
Hidden MMR and the Rank Tier Most Players Actually See
Most modern competitive titles split the raw matchmaking rating, or MMR, from the cosmetic rank shown on a profile. The hidden number drives fair team building, while the visible rank is a friendly translation, usually a Bronze-through-Challenger tier, that makes the climb feel legible. A Gold player can absolutely face Diamond opponents, because the game reads MMR, not the badge.
This split is why placement matches, streaks, and premade squads feel so uneven on the surface. The hidden number absorbs every signal the system has, including recent opponents, party size, and recent performance, while the visible tier only updates after a few matches. The two drift in and out of sync, and the gap between them explains most of the “why is my rank stuck” frustration on the ladder. Watching how fast your badge moves relative to your wins is the clearest signal that those two numbers are no longer aligned.
Smurfing and the hidden number
Behind every ban wave, fresh account flag, and matchmaking shuffle sits a statistical layer developers built specifically to expose smurf accounts. A returning high-skill player or a flagged alternate account can land in harder lobbies before the visible rank catches up, which shields the bulk of the player base from a long string of unfair matches. The cosmetic tier eventually catches up, but the matchmaking engine leans on the hidden number long before the badge does.
Warning: a new account that wins its first ten games is not “stuck in placements” forever. The system is hunting for a level, and your gains will look strange until MMR and rank settle into the same neighborhood.
Elo Derivatives Built for Modern Esports
In 1958, Arpad Elo wrote the math for head-to-head chess matches, and the formulas behind today’s team-based esports have had to be re-engineered from scratch. Three derivatives cover most of the ground, and each keeps the predictive DNA of the original while solving a problem Elo was never built to handle.
| System | Creator / Year | Main Improvement Over Classic Elo | Common Use Cases |
|---|---|---|---|
| Classic Elo | Arpad Elo, 1960 | Predictive pairings and zero-sum updates | Chess, simple ladders, early online play |
| Glicko / Glicko-2 | Mark Glickman, 1995 / 2013 | Adds rating deviation and volatility to weight inactivity | Chess.com servers, online play with long breaks |
| TrueSkill / TrueSkill 2 | Microsoft Research, 2005 | Handles teams of unequal size and free-for-all lobbies | Xbox Live, Halo, Halo Infinite ranked |
Glicko and Glicko-2 added a rating deviation and a volatility measure so a player who hasn’t logged in for three months is trusted less than someone who plays every night. That fix tackled one of classic Elo’s biggest weaknesses: treating a cold account and an active one as equally reliable. Mark Glickman, a statistician, published the upgrade in 1995 and refined it in 2013.
TrueSkill, developed by Microsoft for Xbox Live, extends the same idea to teams and free-for-all lobbies, letting one rating cover up to eight players at once. League of Legends, Dota 2, Counter-Strike 2, Valorant, Rocket League, and Chess.com all run Elo-derived backends under the hood, even when the surface rank looks nothing like a number. FACEIT uses a similar ELO-style ladder for its CS2 hubs.
Reading Your Own Rating Without Misreading It
Numbers mean different things in different games, and even the same game shifts over time. A chess rating of 1200 to 1600 marks a serious club player, 2000 is expert level, and anything above 2700 is grandmaster territory. In Valorant, Immortal sits in a much smaller population than a Dota 2 Ancient, and a Rocket League Grand Champion is not directly comparable to either. Context outweighs the digits for your own progress.
Across competitive scenes, ranks do not translate one-to-one, since each game draws from a different player pool with its own distribution. Win and loss streaks can look like momentum, but they’re really the system correcting an outdated estimate of your skill. A loss after a hot streak often hurts more than expected, because the system raised its prediction of you, and the upset now counts harder against that higher expectation.
How to actually use your rating
Treat the rating as a running prediction of your next game rather than a score to protect, and the climb becomes easier to read. A stable rating means the system is reasonably confident in your level, while big swings mean it’s still learning, or you just produced a result it didn’t expect. Aim to feed it clean, steady games rather than desperate win streaks; the math rewards consistency more than heroics.
Treating your rating as a living signal rather than a trophy reframes how every match should feel going forward.
- Trust the process, not the streak. A five-win run after placement matches usually means the system found you, not that you’ve “broken out.”
- Read losses as data, not damage. A 30-point loss after a big upset is the math working as designed, not a punishment.
- Compare within a game, not across them. Your Valorant rank and your Rocket League rank are not on the same scale, so don’t cross-shop them.
- Watch the hidden MMR, not just the badge. A visible rank that moves slower than your wins suggests is usually catching up to a higher MMR.
- Protect placement games. Early K-factor is high, and the first ten matches shape the system more than the next hundred.
Bottom Line
Elo is a forecasting tool dressed up as a score. The number on your profile is the system’s best guess of who you will beat and who will beat you, and every match moves that guess closer to reality. Once you read it as a prediction instead of a prize, the uneven gains, the slow climbs, and the surprise losses start to feel less like a game fighting you and more like the math doing its job.
FAQ
What is a good Elo rating in competitive gaming?
A good Elo depends on the game and the population you’re playing in. In chess, 1500 marks a competent tournament player, while 2000 is expert and 2700+ is grandmaster level. In online shooters and MOBAs, the top one to three percent of players usually sits in the highest visible tier, and anything above that line is genuinely elite for the platform.
Why does my Elo rating change after winning or losing?
The system compares your actual result to the expected outcome based on the rating gap. A higher-rated player who wins gains a small number because the win was expected. A lower-rated player who wins earns a much larger bonus because the system underestimated them. The change is zero-sum, so the points one player gains, the other loses.
How long does it take to get an accurate Elo rating?
Most games need around 20 to 50 matches to settle a new account onto a stable number, with placement matches carrying the heaviest weight. K-factor stays high during this window so the system can move quickly. After that, your gains and losses shrink, and the rating only drifts when your results consistently beat the system’s expectations.
Is Elo the same as MMR in games?
Elo is the math; MMR (Matchmaking Rating) is the hidden number a game uses internally. Most modern titles run an Elo-derived engine for matchmaking and translate that hidden number into a visible tier, like Bronze through Radiant. So MMR is the Elo-style score behind the scenes, and your badge is the cosmetic version the system shows you.
Which games use the Elo rating system?
Chess.com and FIDE use Elo directly, and most major esports titles run Elo-derived backends. League of Legends, Dota 2, Counter-Strike 2, Valorant, Rocket League, and Halo all use modified versions of the math, often paired with Glicko-2 or TrueSkill to handle teams, breaks, and free-for-all lobbies.
How do professional esports players qualify using Elo?
Many pro circuits use Elo-derived ratings for tournament seeding, league invitations, and open-qualifier cutoffs. Players need a sustained high MMR over hundreds of matches, not a single hot streak, to clear the rating thresholds leagues and tournament organizers set for invite consideration. A consistent track record is what the system rewards, which is why pro careers are measured in seasons, not weeks.
