How Does Elo Rating Work? Points vs Ratings in Quiz Games

You came second in a four-player Compete room and your rating went up. Your friend came second in the next game and theirs went down. Same finish, opposite result.

Rigged? Nope. That is Elo doing exactly what it was built to do.

This guide answers how does Elo rating work in plain words, with real numbers. You will see the formula, why upsets and new players move more, how Guess the Celeb's Compete mode turns a room of up to 10 players into Elo, and what the tiers mean. Then we look at the other kind of leaderboard: boards that count effort instead of skill, and why that is sometimes the better design.

Key takeaways

• An Elo rating is a prediction. The gap between two ratings says who should win, and the result moves both ratings by the size of the surprise.
• Upsets move more points than expected results. Beating a stronger player pays more than beating a weaker one.
• The K-factor sets the maximum swing. New players get a bigger K so their rating finds its level fast.
• Compete treats every opponent in a room as a one-on-one match, then divides by the number of opponents, so a 10-player game moves you about as much as a duel.
• In Compete, everyone starts at 1000 and climbs through Rookie, Regular, Sharp, Elite and Legend.
• Ratings only mean something inside their own pool. A 1300 in Compete is not a 1300 in chess.
• Use a rating when you want to know who is strongest. Count effort when you want more of a useful behaviour.

In this guide

• What is Elo rating, and how does it work?
• How does Elo rating work? The formula in plain numbers
• K-factor: why new players swing more
• How Compete turns a 10-player room into Elo
• Elo rating tiers in Compete
• Is a 1000 rating good?
• Elo, Glicko and other rating systems
• What moves your Compete rating (and what doesn't)
• Leaderboard points vs rating: which board measures what
• When a leaderboard should count effort
• Scorecards: rating places, not people
• How to keep an effort board honest
• Common mistakes when reading a rating
• Troubleshooting: my rating did something odd
• FAQ

What is Elo rating, and how does it work?

An Elo rating is a prediction. Before a match, the gap between two ratings tells you how likely each player is to win. After the match, the system compares what happened with what it predicted, and moves both ratings by the size of the surprise.

That is the whole idea. A rating is not a score you collect. It is the system's current best guess at how strong you are, updated after every game.

The name is not an acronym. The system is named after Arpad Elo, a Hungarian-American chess master and physics professor, so it is written "Elo", not "ELO". According to Wikipedia's history of the Elo system, the US Chess Federation implemented his method in 1960 and FIDE, the world chess federation, adopted it in 1970.

Elo rating in games now goes far beyond chess. The same article lists adaptations for team sports, board games and esports. Quiz games are a natural fit too: every round has a winner, a loser and, now and then, a dead heat.

How does Elo rating work? The formula in plain numbers

The prediction is called the expected score. It runs from 0 (certain loss) to 1 (certain win). Here is how it changes with the rating gap:

The formula is E = 1 / (1 + 10^((opponent - you) / 400)). You don't need to memorise it. Remember one thing: every 400 points of gap makes the favourite ten times more likely to win in odds terms.

After the game, your new rating is:

new rating = old rating + K × (actual score - expected score)

A win counts as 1, a loss as 0 and a draw as 0.5. K is the maximum swing, covered in the next section.

Upsets move more points

Say you are on 1000 and you beat someone on 1150. Your expected score was about 0.30, so you overperformed by 0.70. That is a big surprise and a big gain. If you had lost, you would have underperformed by only 0.30, a small surprise and a small loss.

With Compete's K of 48 for new players, that works out to +34 for the upset win or -14 for the expected loss. Beating someone far below you does the opposite: a tiny gain, and a painful loss if you slip.

Draws still move ratings

A draw is not "nothing happened". If a 1000 and an 1100 draw, the 1000 scored 0.5 against an expectation of about 0.36, so they gain a little and the 1100 loses a little. At K = 32 that is +4 and -4. In Compete, a draw means two players finished a game on exactly equal points.

K-factor: why new players swing more

K controls how far one game can move you. A new player's rating is mostly a guess, so a high K lets it find the right level quickly. An established rating has lots of evidence behind it, so a lower K keeps it steady.

Chess does this too. Under the FIDE Rating Regulations applied from 1 March 2024, K is 40 for players new to the list until they have completed at least 30 games, 40 for juniors until the end of the year they turn 18 while they stay under 2300, 20 for players under 2400, and 10 once a player's published rating reaches 2400.

Compete uses its own, faster-moving scale:

The game's own code gives the reason in one line: new players move fast so the number means something after a few games, and veterans settle.

How Compete turns a 10-player room into Elo

Elo was designed for one-on-one games, so how does Elo rating work when up to 10 players share a Compete room? The trick is to treat every game as a set of head-to-head matches:

1. Every opponent is a match. In a four-player room, you have three matches: one against each other player.
2. Points decide each match. If you finished with more points than someone, you beat them. Fewer, you lost. Exactly equal points is a draw.
3. Expected scores use both ratings, the same formula as chess.
4. The total is divided by the number of opponents, then multiplied by your K and rounded to a whole number.

That last step matters. Without it, a 10-player game could move your rating nine times as far as a 1v1. With it, a big room and a duel move you by similar amounts.

Worked example: a four-player room

You are on 1000 with 20 games played (K = 32). The room: player A on 1100 with 60 games, player B on 1000 with 5 games, and player C on 950 with 30 games. A finishes first, you second, B third, C fourth.

Your three matches:

• A outscores you: actual 0, expected 0.36. Result: -0.36.
• You outscore B: actual 1, expected 0.50. Result: +0.50.
• You outscore C: actual 1, expected 0.57. Result: +0.43.

Total: +0.57. Divided by 3 opponents and multiplied by 32: +6. Here is the whole room:

Notice that A won the room but gained only 8. A was expected to win most of those matches, so winning them was not much of a surprise.

Why a 10-player win counts like a 1v1

Take ten brand-new players, all on 1000 and all on K = 48. The winner beats nine equal opponents, so every match is worth +0.5. Nine times 0.5, divided by nine, times 48: +24. Last place gets -24. That is exactly what a 1v1 between two new players on 1000 gives. Big rooms are not a shortcut to a big rating.

When second place loses rating

A player on 1300 with lots of games behind them (K = 24) finishes second in a room with players on 1000, 1100 and 1050. They beat the 1100 and the 1050, but lose to the 1000. That one loss was a big surprise, and the two wins were expected. Net result: -3. Second place, and the rating still dipped. The 1000 who won the room, on K = 32, gains 22.

Elo rating tiers in Compete

Everyone starts at 1000. Your tier comes from your rating:

Your rating and tier show on the end screen, your profile, shareable result pages and the Compete leaderboard, which is sorted by rating. The live Compete page puts the climbing rule simply: beat higher-rated players to climb faster.

The gaps between tiers are not equal on purpose. Regular is 50 points above the start, a handful of good games. Legend is 500 points up, which means beating strong players again and again.

Is a 1000 rating good?

In Compete, 1000 is where everyone starts, so it means "no evidence yet". After a few games, a rating near 1000 means you are winning about as often as you lose against the players you meet. Climb to 1050 and you are a Regular.

Whether a number is "good" depends entirely on the pool. Elo ratings are comparative only and valid only within the pool where they were calculated, as the Wikipedia article puts it. Two quick comparisons show why:

• FIDE only publishes ratings of 1400 and above, so 1000 is not even on the official chess list.
• Lichess starts every new player at 1500 on a different system, Glicko-2, according to its own FAQ.

So "is Elo rating 1000 good" has no single answer. In Compete it is the starting line. Elsewhere it can mean a beginner, a typical player or something else entirely.

Elo, Glicko and other rating systems

Elo has a known weakness: it treats every rating as equally certain. A player with 3 games and a player with 300 games on the same number look identical. The K-factor is a rough patch for that.

Glicko, developed by Mark Glickman, adds a second number: how confident the system is in your rating. Lichess's FAQ explains that its Glicko-2 ratings show a question mark while that uncertainty is high, which marks the rating as provisional. FIDE, by contrast, still uses Elo.

Many online games pair players using some kind of skill rating. The details vary from game to game, so treat a number from one game as meaningless in another.

Compete keeps it simple: classic Elo, a stepped K-factor instead of an uncertainty number, and pairwise matches for multiplayer rooms. That is easy to explain on an end screen, which matters in a party game.

What moves your Compete rating (and what doesn't)

• Speed counts, indirectly. Faster right answers earn more points when the host leaves the speed bonus on, and points decide who beat whom. A correct answer in two seconds can flip a head-to-head that a slow correct answer would lose.
• Who you play counts. Beating stronger players pays more. Rooms full of players far below you risk more than they offer.
• Your first 10 games count most. K is highest then, so early results move you fastest. Don't panic about a bad start; it settles.
• Your finishing place doesn't count directly. Only the head-to-heads do. That is why second place can go either way.
• Games against the House don't count. If you wait alone in a public lobby, the site can seat a bot called the House so you can play. Those games are unrated.

If you want warm-up practice before a rated game, a 10-face easy round you can play without signing in uses the same speed-based points. The hard celebrity quiz is closer to the pressure of a strong room. When you are ready, open live multiplayer for up to 10 players in Compete; it needs a free Google sign-in.

Leaderboard points vs rating: which board measures what

Not every leaderboard should be a rating. Guess the Celeb runs both kinds. The Compete board is a rating ladder. The GuessTheCeleb leaderboard for today, this week and all time is a points race: right answers and speed add up, and ties go to the faster total time.

A rating ladder answers "who is better?". A points board answers "who did the most, or did it fastest, in this window?". An effort board answers "who helped most?". All three are fair. They just answer different questions.

The daily is a good example of why points fit some games better. Today's three-face daily celebrity round, which needs a free Google sign-in, is a habit, not a tournament. A rating would barely move on three faces a day, while points and streaks reward coming back.

When a leaderboard should count effort

Some activities have no opponent at all: a step challenge, a reading log, a neighbourhood clean-up. Nobody is beating anybody, so Elo makes no sense. There is no match to predict and no surprise to measure. What the organiser wants is more of the activity, so the board should count exactly that.

For a real example, take a civic app's citizen leaderboard: the one in Potholes.app, an independent map of road hazards for Pune and Pimpri-Chinchwad. It is not a PMC or government app, and it does not file complaints. Residents pin a photo of a hazard, and other residents check it. The About page explains how neighbours confirm or close each Potholes report: the map stays public so people can confirm what is still there and what has been repaired.

The scoring rewards both halves of that loop. In the Android build 1.0.10 (the app is in closed testing), the scorecard screen spells out the points: reports earn 10, repairs 15, confirmations 2, and backing a neighbour's report 1. Points add up to seven citizen levels, from New Citizen to Civic Champion, and there are ten badges. The scoring may change in later versions, but the design is the lesson. Checking someone else's report earns points too, so the board rewards verifying as well as posting.

Scorecards: rating places, not people

Potholes also rates something that is not a person: wards. Its ward scorecard combines two measures, and the legend on screen states the weights: score = 60 percent resolution rate + 40 percent fix speed, against a 30-day target.

Fix speed runs from 100 for a same-day repair down to 0 for an average fix time of 30 days or more. The bands are 70 and above "Doing well", 40 to 69 "Mixed", and below 40 "Failing citizens".

Here is a made-up example to show how the weights interact:

Two very different wards land on the same score. One fixes more but slowly; the other fixes fewer but fast. That is the point of a weighted scorecard: it makes the trade-off visible, and the 60/40 split says resolution matters a bit more than speed.

Notice what feeds it. A ward score is only as good as the reports and checks behind it, which is why effort boards and scorecards work together. Residents who run a repeat neighbourhood road walk in Pune and mark each pin "still there" or "fixed" are exactly the effort the citizen board rewards.

How to keep an effort board honest

Effort boards have one big weakness: people do the cheap version of the effort. Good designs push back in four ways.

Count the first attempt, not the replay

Guess the Celeb's points boards count only your first guess on each card, so replaying a card you have already seen cannot farm points.

Make the useful action pay more than the lazy one

On Potholes, a report earns 10 and a repair 15, while a confirmation earns 2. That keeps the big numbers on real reports and real outcomes, while still paying something for the checking work.

Write the rules down and limit spam

The Potholes community guidelines set rules against duplicate pins and invented hazards, and say reports, comments, votes and posts are rate-limited. Repeat spam can close an account. An effort board without rules like these turns into a spam board.

Tell people what the score is built from

People trust a score they can understand. Potholes' privacy policy on its civic reputation score describes it as derived from in-app civic actions, and the scorecard screen lists the points per action. A hidden formula invites suspicion; a published one invites the behaviour you want.

Common mistakes when reading a rating

• Comparing across pools. A Compete 1300, a chess 1300 and a Lichess 1300 measure different things.
• Reading finishing place as the result. In multiplayer Elo, only the head-to-heads count.
• Judging after three games. Early ratings swing hard by design. Give it 10 games.
• Chasing big rooms for big gains. The division by opponents makes a 10-player win worth about the same as a duel.
• Using a rating for a daily habit. Three faces a day suits points and streaks, not Elo.
• Using points for "who is best". A points race rewards volume. If you want strength, use a rating.

Troubleshooting: my rating did something odd

Most "how does Elo rating work?" questions start right here, with a number that moved the wrong way.

• I finished second and lost points. You probably lost to someone rated well below you. That upset outweighed your expected wins.
• I won the room and barely gained. You were the favourite against most of the room, so winning was expected.
• My rating did not change at all. Check whether the House sat in. Games with the House are unrated. Small rounding can also land a tiny change on zero.
• My rating jumped 20 or more in one game. You are probably still in your first 10 games, where K is 48.
• Two of us tied on points. Equal points count as a draw between you, worth 0.5 each against your expectations.
• My friend and I got different changes for the same result. You have different ratings and different game counts, so different expectations and different K values.

FAQ

What does Elo stand for in games?

Nothing. Elo is a surname, not an acronym. The system is named after Arpad Elo, the chess master and physics professor who designed it for chess, which is why it is usually written "Elo" rather than "ELO". Games and apps that say "Elo" usually mean a rating that rises when you beat stronger players and falls when you lose to weaker ones, even if their exact maths differs from the original.

How does Elo score work in a multiplayer game?

The common approach, and the one Compete uses, is to split the game into head-to-head matches. You are compared with every other player: more points is a win, fewer is a loss, equal is a draw. Each match uses the normal expected-score formula. The results are added up, divided by the number of opponents and multiplied by your K-factor, so a big room moves you about as much as a 1v1.

Why did I lose rating after finishing second?

Because the player who beat you was rated well below you, and that surprise outweighed your expected wins against the others. In the worked example above, a 1300 player finishing second in a room of lower-rated players drops 3 points. Finishing place is not the input. The head-to-head results against each opponent's rating are.

Does winning a 10-player game count more than winning a 1v1?

Not really. Your total is divided by the number of opponents. Ten new players on 1000 give the winner +24, exactly what a new player gets for winning a 1v1 against another 1000. What earns more is beating stronger players, not beating more of them.

Can my Compete rating fall below 1000?

Yes. 1000 is the starting point, not a floor, so losing to lower-rated players can take you under it. The game code does set a floor far below that, at 100, so a rough run cannot send a rating to zero. A rating under 1000 still has the tier Rookie.

Can I use Elo for my own quiz league?

Yes, and it works well for regular teams. Keep a spreadsheet with each team's rating, start everyone at 1000, treat each quiz night as head-to-head matches between teams, and use a K of about 32. For a casual night where you just want people to come back, a points table and a "most improved" prize may suit better. How to Host a Quiz Night With Friends covers keeping a regular night going.

Where to go next

The fastest way to answer "how does Elo rating work?" for yourself is to watch your own number move. Start a rated Compete room with friends, check the end screen after each game, and see which head-to-heads moved you. Between games, our guides for players and hosts cover quiz nights, picture rounds and more.

If you are designing a board for something without opponents, like a neighbourhood cleanup or tree-planting drive in Pune, borrow the effort-board ideas above: count the useful action, pay more for the harder version, publish the formula and block the spam.