The Question Mark Is Part of the Rating

A chess rating looks like a fact.

1167.

Four digits. No visible hesitation. It has the clean authority of a measurement taken by a machine that definitely remembered to calibrate itself.

Then Lichess places a question mark after it.

1167?

That punctuation changes the emotional reading. The number no longer seems official. It looks unfinished, suspicious, perhaps embarrassed to be seen in public.

The question mark is not an insult and it is not a warning that the account has done something wrong. It is the rating system being more honest than the number normally allows.

The rating is an estimate.

The question mark tells us the estimate is still uncertain.

What a rating is trying to do

A rating system watches results and estimates playing strength. It does not know how well I calculate, whether I understand rook endings, or why I keep moving quickly in classical games. It receives a much smaller report:

• who played;
• who won, lost, or drew; and
• what the system believed about both players before the game.

From that, it updates its estimate.

The useful word is relative. A rating does not measure a universal quantity like temperature or distance. It locates a player inside a particular pool of players under particular rules.

That is why a 1500 on Lichess, a 1500 on Chess.com, and a 1500 FIDE rating do not mean the same thing. The pools differ. The time controls differ. The starting conditions differ. The update formulas differ. Even two services using closely related mathematics can produce different scales because different people are supplying the results.

Lichess makes this point directly: ratings should be compared inside the same rating pool, not treated as absolute numbers.

Asking whether a Lichess 1500 “is really” a FIDE 1500 is therefore a little like asking whether fifteenth place in one marathon is the same speed as fifteenth place in another. The positions look identical. The fields and performances may not be.

The names behind the numbers

“Elo” has become casual shorthand for almost any competitive rating. It is also the name of a specific family of calculations. Not every number that rises after a win is Elo.

Elo

Arpad Elo’s system estimates each player with a rating number. The difference between two ratings implies an expected result. A player who scores better than expected gains points; a player who scores worse loses them. A development coefficient, usually called the K-factor, determines how strongly a result moves the rating.

Elo’s great advantages are clarity and practicality. The basic calculation can be explained, checked, and administered without a giant technical system. FIDE adopted Elo decades ago and still uses an Elo-based method for its official rating lists.

The limitation is that the displayed rating does not explicitly tell us how uncertain the estimate is. A player with five recent games and a player with five hundred can both be shown as a single number, even though one estimate is much less settled.

Glicko

Mark Glickman’s Glicko system extends Elo by tracking rating deviation, usually abbreviated RD.

The rating remains the system’s best estimate of strength. RD describes how uncertain that estimate is. Low RD means the system has a narrow idea of where the player’s strength lies. High RD means the plausible range is wide.

This changes how updates behave. A new or long-inactive player can move quickly because the system admits that its old estimate is weak. A frequently active player with consistent results usually moves less because the system has accumulated evidence.

Chess.com says it uses the Glicko system and considers both players’ rating confidence when calculating changes.

Glicko-2

Glicko-2 adds volatility to rating and RD. Volatility describes the degree of expected fluctuation in a player’s rating: steady results imply one pattern, while unusually erratic results can imply another.

The full Glicko-2 model therefore keeps three related values:

1. the rating estimate;
2. the rating deviation, describing uncertainty; and
3. volatility, describing expected movement in the estimate.

Glickman’s own explanation recommends thinking in intervals rather than pretending the rating alone captures everything.

Lichess uses Glicko-2.

TrueSkill

Microsoft’s TrueSkill approaches the same problem with multiplayer games and teams in mind. It represents skill with a mean and a standard deviation—a best estimate and uncertainty—and can update several players from one match.

That matters in games such as Halo, where a result might involve two teams rather than one White player and one Black player. Microsoft describes TrueSkill as both a ranking and matchmaking system and uses the uncertainty value when deciding how confidently it knows a player.

The shared idea is more important than the brand name: modern systems often store both what they currently think and how sure they are.

Why Lichess shows a question mark

On Lichess, the question mark means the rating is provisional.

The precise rule is not “play ten games” or “win once against an established player.” Lichess displays the question mark when Glicko-2 rating deviation is greater than 110.

That is why nobody can promise an exact number of games before it disappears. Each result supplies evidence, but the amount and quality of that evidence vary. Games against opponents whose ratings are themselves uncertain tell the system something different from games against established opponents near the estimated strength.

Every rating category also has its own evidence. Playing thirty blitz games does not establish my classical rating. The skills overlap, but the pools are separate because performance changes with the clock.

The question mark can also return after inactivity. Glicko-2 allows RD to grow when a player does not compete. This is not decay in the player’s rating. The system is not declaring that absence made the player worse. It is declaring that the old evidence has become less reliable: perhaps the player studied for a year, perhaps stopped playing entirely, perhaps returned with different strength.

This is exactly the situation uncertainty is for.

Why not hide the number until it is stable?

Because the provisional estimate is already useful.

The system needs somewhere to begin matchmaking. It can use the current estimate while allowing large corrections. A new player who is much stronger or weaker than the starting guess moves toward an appropriate pool quickly.

Showing the number also lets the player see that process. The question mark supplies the missing sentence:

This is our current estimate, but do not build a monument to it yet.

Why are the early swings so large?

Because small adjustments would pretend to know more than the system knows.

If the possible skill range is wide, a decisive result against an established opponent is strong evidence. Moving a provisional rating by a hundred points can feel violent, but refusing to move it would leave the player in mismatched games for longer.

Fast correction is not instability for its own sake. It is how the system searches for the right neighborhood.

Lichess, Chess.com, and FIDE

The three most common numbers in online chess conversation answer related but different questions.

Under the current FIDE standard rating regulations, a new rating is published after at least five games against rated opponents, provided the calculated rating reaches the 1400 floor. Results may be pooled across rating periods. FIDE publishes an official list monthly rather than presenting every game as an immediately settled public rating.

This is a different answer to the provisional problem. Lichess publishes an estimate and attaches a question mark. FIDE waits until its publication conditions are satisfied.

Neither choice turns the number into an absolute measure across pools.

Why the numbers drift apart

Suppose Lichess and Chess.com used identical formulas tomorrow. Their ratings would still not become interchangeable.

Each service has its own:

• player population;
• starting-rating policy;
• mix of casual and serious competitors;
• distribution across time controls;
• frequency of play;
• approach to inactive accounts; and
• history of adjustments, new accounts, and removed accounts.

A rating scale is an ecosystem. Copying the ruler does not copy the population being measured.

Rating, rank, tier, and title are different things

Competitive games often place several systems on top of one another.

A rating is a numerical estimate used to compare performance.

A rank is an ordering: first, second, 10,482nd.

A tier is a named bucket such as Silver, Gold, or Diamond.

A title is an achievement granted under separate rules, such as FIDE Master or Grandmaster. A title may require rating thresholds and tournament results, but it is not simply the live rating wearing a name tag.

These layers can move together, but they do not have to be the same system.

That distinction explains why competitive video games sometimes show one thing while matchmaking with another.

World of Warcraft: the number that makes the match and the number that earns the reward

World of Warcraft Arena has long distinguished matchmaking rating, or MMR, from visible competitive progress.

Blizzard described Arena opponents as being selected by the average MMR of the queued party. Rated PvP has also used named bands such as Combatant, Challenger, Rival, Duelist, and Elite for visible progress and rewards.

Those two layers have different jobs:

• MMR can react in the background to assemble competitive matches.
• Visible rating and named ranks give the season a legible progression path.

The advantage is flexibility. The matchmaker can chase fair games without making every internal correction feel like the player has lost a badge.

The disadvantage is confusion. If the hidden estimate and visible progress separate, players see gains and losses that feel arbitrary. They know the result. They cannot see the complete reason for the update.

League of Legends: MMR underneath, tiers on top

League of Legends makes the separation even more visible. Players talk in Iron, Bronze, Silver, Gold, and the tiers above them. Progress arrives through ranks and League Points. Matchmaking operates around an underlying MMR.

The tier is not the rating algorithm. It is a human-facing progression system placed over the matchmaker’s estimate.

Riot has acknowledged that the hidden relationship between visible rank and MMR creates confusion. The same post describes goals beyond raw skill matching: queue time, role assignment, premade balance, transparency, progression satisfaction, and rewards.

Chess matchmaking is comparatively clean. One player, one opponent, alternating colors, no jungle role, no five-person party, no question about which teammate was autofilled. Even then, rating is only one constraint. A platform may also consider time control, acceptable rating range, color history, tournament score, rematches, and how long someone has waited.

Team games expose the general problem: the statistically closest match is not always the best available match, and the best possible match is not useful if it takes forty minutes to assemble.

What ratings do besides matchmaking

Fair pairings are the most immediate use, but ratings become infrastructure for an entire competitive environment.

Seeding and pairings

Ratings help tournament systems avoid random early mismatches, create sections, seed brackets, and choose reasonable pairings inside larger events.

Leaderboards

A rating can become an ordering system. Leaderboards turn estimates into public competition, provided the platform also defines activity and eligibility requirements. Lichess, for example, requires recent play and tighter rating deviation for its top-ten leaderboards than it requires merely to remove the provisional question mark.

Entry requirements and qualification

Events can set rating floors or ceilings. Invitational systems can use ratings as one qualification path. Amateur sections can prevent experienced players from entering lower brackets. These uses make rating accuracy matter even when no automatic opponent is being selected.

Progress and feedback

A long-term rating trend can show that results are changing. It cannot diagnose why. The number may tell me that I am scoring better against the pool; it cannot tell me whether the cause is tactical vision, endgame technique, or finally taking more than twelve seconds on move fifteen.

Rewards and recognition

Games attach titles, cosmetics, seasonal rewards, and access to visible rating bands because a raw probability estimate is not much of a celebration. Tiers translate a continuous number into milestones people can understand and pursue.

Measuring the system itself

A rating system can be judged by prediction. If players with a stated expected score consistently perform very differently, the model or pool may need attention. Match quality, queue time, smurf placement, inactivity, and population size all become operational questions built around the estimate.

The tradeoffs

There is no rating system that maximizes every useful property at once.

Simple and transparent versus adaptive

Elo-style calculations are comparatively easy to explain and audit. Systems that explicitly model uncertainty can place new and returning players faster, but the updates become harder to reproduce by hand.

Visible versus hidden

A visible rating gives players understandable feedback and allows them to verify changes. A hidden MMR gives a matchmaker freedom to adjust without turning every experiment into public drama. Hiding the important number can also damage trust when outcomes do not match visible progress.

Stable versus responsive

Small updates make an established rating feel meaningful. They can be too slow for a rapidly improving beginner or a strong player on a new account. Large updates find the correct pool faster but make individual games feel enormous.

Glicko’s answer is not to choose one speed forever. It changes the speed according to uncertainty.

Precise-looking versus honest

One number is clean. A number plus an uncertainty interval is more truthful and harder to love.

The question mark is the smallest possible uncertainty disclosure. It does not explain RD, volatility, rating periods, or confidence. It simply interrupts false precision.

How I want to read my rating

I already prefer to hide ratings while I play. The board supplies enough information without asking a four-digit estimate to participate in move selection.

After the game, the number can return to its proper job.

It can help Lichess find the next opponent. It can place tournament entrants in roughly sensible company. Over many games, it can show whether my results against the pool are changing.

It cannot certify the quality of a single game. It cannot be compared directly with a number from another pool. It cannot tell me what to study tomorrow. It is not a title, a personality test, or a permanent verdict.

And while the question mark remains, the system is saying something unusually modest:

We have an estimate. We are still gathering evidence.

That is not a defect in the rating.

That is part of the rating.