How ratings work: CSR and TSR

ChesStresS has two ratings. CSR – ChesStresS Rating measures your results in rated online chess games, one rating per speed. TSR – Turbo StresS Rating measures how well you read your team in rated Turbo Live events. Both are our own ratings: CSR and TSR are not FIDE ratings and are not linked to FIDE, Lichess, Chess.com or any other organisation.

Contents: How CSR works · Worked example · Leaderboards · How TSR works · Turbo Live entry fees, top-ups and badges · Fair-play corrections · Not a FIDE rating · FAQ

Every number in the tables on this page is calculated by the game server's own code (worker/scripts/rating-example.mjs runs it and writes the tables), so the page and the game cannot disagree.

1. How CSR works

CSR uses the Glicko-2 rating system, published by Professor Mark Glickman (The Glicko-2 System for Rating Players in Head-to-Head Competition, glicko.net). Glicko-2 keeps three numbers for you in each speed:

• Rating – the strength estimate, the number you see (for example CSR 1843). A new player starts at 1500.
• Rating deviation (RD) – how sure the system is about that number, shown as “±”. CSR 1843 ±60 means your strength is most likely within about two RDs of 1843 (here, roughly 1723 to 1963). A new player starts at ±350: nothing is known yet.
• Volatility – how much your results jump around compared with what your rating predicts. It is kept by the system and is not shown; it changes slowly.

Why Glicko-2 and not the classic Elo system

The classic Elo system (used by FIDE in a simpler form) gives every player one number and moves it by a fixed amount per game. Glicko-2 adds the deviation, which fixes three problems:

• Uncertainty. When little is known about you (high RD) your rating moves a lot, so it finds your level quickly. When a lot is known (low RD) it moves only a little. A game against an uncertain opponent also counts for less than a game against a well-known one.
• Inactivity. If you stop playing, the system becomes less sure about you: your RD grows with time, so your first games after a break move your rating more.
• Provisional ratings. While your RD is high the rating is marked “?” and is not listed on leaderboards.

Our parameters

Categories: which rating a game changes

CSR has five separate ratings: Bullet, Blitz, Rapid, Classical, Correspondence. A game's category comes from its time control: the estimated length is the base time in seconds plus 40 times the increment (an average of 40 moves). Under 180 seconds is Bullet, under 480 seconds Blitz, up to 1,500 seconds Rapid, longer Classical. Correspondence games (days per move) are Correspondence.

What counts for CSR

• Counts: rated online games between two accounts, finished with a result: rated games from the lobby, direct challenges marked rated, and tournament games.
• Does not count: casual games, Chess960, games without a time limit (“unlimited”), games against a guest, “Challenge a friend” link games, aborted games, Turbo Live (it has TSR), and games against the computer.
• Your rating changes after every rated game (one game is one rating period). Both players' new ratings are worked out from the ratings both had before the game.
• White's first move. White scores a little more than Black at equal strength, so CSR works out the expected result as if White were 35 points stronger. A draw with White against an equal opponent therefore costs you a little, and a draw with Black gains a little.

Inactivity

For every day without a rated game in a category, your RD grows: new RD = √(RD² + (volatility × 173.7178)² × days), never above 350. With the usual volatility of 0.06 that is about 10.4 points times the square root of the days. The growth is applied when you next play a rated game; the Account tab and the leaderboards show your RD as it was after your last game.

Provisional ratings (“?”)

A rating is provisional while its RD is above 110. Starting from ±350 that takes about 10 games against established players (1500 ±60), or 14 games against players who are still uncertain themselves (1500 ±200). A long break can make a rating provisional again.

Starting from your other ratings

A new account starts every category at 1500 ±350. But once you have at least 5 rated games in a category, a category you have not played yet starts from what is already known about you: the average of your other ratings with at least 5 games, weighted by how close the speeds are (neighbouring speeds count fully, two steps away half, further a quarter; Correspondence sits next to Classical) and by how sure each rating is (1/RD²). The start has RD ±250, so it is provisional (“?”) and settles quickly, but a strong Blitz player does not have to beat a long line of beginners in Bullet first, and they do not lose points to him. Until your first rated game in that category the start follows your other ratings; after it, the category is an ordinary rating.

Rating floor

Losing on purpose to get a low rating (“sandbagging”) breaks fair pairing, so each CSR has a floor. Once a category is established (RD 110 or lower after a game), its highest rating so far (your peak) is kept, and the rating can never fall below a floor of 400 points below your peak, rounded down to a hundred, and never below 100. A loss that would take you lower stops at the floor. The Account tab shows the floor next to the rating. TSR has no floor (agreement cannot be lost on purpose in a way that helps you). Losing on purpose also stays against the rules and is looked for by our fair-play checks.

Practice ratings against the computer

Your timed games against the computer change a practice rating kept on your device only (Profile › Practice ratings & saved games), using the same formulas with the computer's strength as the opponent. It is labelled “vs computer, not CSR”: the server does not use it, it does not appear on leaderboards and it never changes your CSR.

2. Worked example (CSR)

Two new players, a 5+0 game (Blitz), White wins. Neither has played any rated game before, so both start at 1500 ±350. White's expected score is a little above one half because of the first-move adjustment.

That is +152 for White and −152 for Black: big steps, because neither rating was known. Both RDs fall, because the game told the system something.

Two established players, a draw. White is 1843 ±60, Black 1700 ±80, both last played yesterday.

White was expected to score about 0.73, so a draw is below expectation and White loses a few points; Black gains a few. The steps are small because both ratings are well known.

3. Leaderboards

• One CSR leaderboard per category and one TSR leaderboard (Profile › Leaderboards; the home screen shows the top 10 of the category of your home time control).
• The top 50 by rating. Only established ratings are listed: RD below 110 and at least one rated game (or event, for TSR).
• Only ordinary player accounts are ranked: family accounts (unlimited tokens, used by the operator's family and for testing) and the operator's own accounts are left out.
• Boards are kept for up to a minute; after a rated game in that category the board is rebuilt within a few seconds.

4. How TSR works BETA

TSR is in beta. Turbo Live is new, and we may still tune TSR's weights (how much the field and the team part count, the votes needed, the draw margin) as real events show how it behaves. Any change will be listed on this page with its date, and TSRs earned before a change are kept, not recalculated.

In Turbo Live every move is a team vote, and your agreement is the share of your team's moves that matched your vote. TSR rates how well you read your team, compared with everyone else in the same event. It uses the same Glicko-2 scale and parameters as CSR (start 1500 ±350, volatility 0.06, τ 0.5, RD floor 45, “?” above 110, RD growth with inactivity), without the White adjustment.

• One update per event. TSR changes once, when a rated event ends (for the Weekly Marathon: when the week ends). Everyone's new TSR is worked out from the TSRs as they were before the event.
• Who is rated. Accounts that cast at least 10 votes in the event, in rated events only (events with an entry fee). Free events are unrated. Guests have no TSR.
• The field. You are compared with every other rated player in the event, as if you had played each of them: higher agreement is a win, lower a loss, and a difference of 1 percentage point or less is a draw. All these comparisons together count as 4 games, shared equally (with 3 opponents each comparison weighs 4/3 of a game; with 100 opponents, 0.04). In very large events you are compared with 200 players picked evenly along the agreement order, the same ones every time, which keeps the calculation fast and fair.
• Your teams. Your team results in games against another human team (wins plus half the draws, divided by the games) count as one more game, against the average TSR and RD of the other rated players.
• The engine and playing alone. When a team has nobody online, Stockfish plays for it. A game with engine moves is practice from its first engine move, and the game says so on the screen: nothing in it counts from then on – no votes, no agreement, no team result – so it gives no points, no badges and no TSR. The engine's strength is fixed and says nothing about the human players. A rated event cannot be started with one player, and an event with fewer than 2 rated players changes nobody's TSR. Practice (offline Turbo, with a simulated crowd) never counts.
• Top-ups never change TSR. A top-up lets you stay in the event, and the game then treats your agreement as if it were on the team line. TSR always uses your real agreement, counted before any top-up.

Worked example (TSR)

A rated 10-minute event with five accounts. Eve cast only 9 votes, so she is not rated, whatever her agreement.

Anna’s results: 3 field wins of weight 1.33 each (4 ÷ 3 opponents), and a team score of 0.67 (2 wins of 3) with weight 1 against the others’ average, 1550 ±93. Ben (65.0%) and Cleo (64.5%) are within 1 point, so their comparison is a draw. Anna was new (±350), so one strong event moves her a long way; Dev, well known at ±70, moves less although he lost every comparison.

5. Turbo Live entry fees, top-ups and badges

• Entry fee. Rated events have an entry fee in tokens. It can be paid with earned or bought tokens (bought tokens are used first). It buys your place in the event and nothing else.
• No token prizes. Turbo Live pays no tokens to anyone. What you play for is your TSR, your place in the standings and ladders, and badges.
• Refunds. Your fee comes back if the event is cancelled (a rated event that has not started 30 minutes after the first entry), if you withdraw before your first game, or if the event ends without you ever being placed in a game. Every charge and refund is in your token history.
• Top-ups. If your agreement falls below the line you may stay in by paying a top-up, from earned tokens only, up to 3 times per event, each dearer than the last. The 15 seconds to decide are a game rule; the game never offers to sell you tokens in that time. If you do not have enough earned tokens, the dialog says so and offers only “Drop out”. The Get tokens screen is closed while you play in an event.
• Free events cost nothing, are open to guests and are unrated. Family accounts enter rated events without paying. Guests can watch rated events.
• Badges are cosmetic, shown on your profile, cannot be traded and have no value: “Turbo Live winner” for first place in a rated timed event with at least 4 rated players, and “Marathon Champion”, “Marathon Top 10” and “Marathon Top 100” for the week (for example “Marathon Top 10 — week 46, 2026”). Places count among ordinary player accounts with at least 10 votes; family and the operator's accounts get no badges.
• Ladders. The Weekly Marathon ladder (this week's top 100 and last week's top 100) and the TSR leaderboard are visible to everyone.

6. Fair-play corrections

Using an engine or other help, sharing an account, losing on purpose or agreeing results are against the rules (terms, section 7). Reports are reviewed by a person (Safety & reporting). If a person decides, after review, that an account cheated (it is closed or marked for fair-play violations), the game corrects the ratings of the players it met, automatically:

• CSR. For every rated game in the last 90 days against that account, a player who lost CSR in that game gets exactly those points back. Points a player gained against it are not taken away.
• TSR. Every rated Turbo Live event of the last 90 days in which that account was rated is worked out again without it, from the event's stored results. A player whose TSR would have been higher gets the difference; nobody loses TSR. (If fewer than 2 rated players are left, the event counts as if it had not changed anyone's TSR.)
• Each correction is shown to the player under Profile › Account › Notices. If the decision is overturned on appeal, the corrections are taken back the same way.
• Tokens are never moved automatically: a token prize won against a cheater is looked at by a person.

Accounts closed for other reasons (abuse, or because the owner is under 13) do not change anyone's rating.

7. CSR is not a FIDE rating

CSR and TSR are calculated only from games and events played in ChesStresS. They are not FIDE ratings, are not reported to FIDE or any federation, and cannot be compared directly with FIDE, national, Lichess or Chess.com ratings: each pool of players has its own scale. Ranks such as “Grandmaster I” are game levels, not titles (About & credits). The strengths of computer opponents (“Stockfish ~1700 Elo”) are Stockfish's own strength settings, not CSR.

8. FAQ

Why did I gain only 3 points for a win? Your gain depends on how surprising the result was and on how sure the system is about you. For example, with White, 1843 ±50 against 1600 ±60, your expected score is 0.828; a win scores 1, so you gain only +3 (to 1846). Beating a stronger player, or a win while your RD is high, moves you much more.

Why is my rating marked “?” Your RD is above 110: the system is not yet sure about your level, because you have played few rated games in that category or have not played for a long time. It is not listed on leaderboards until the “?” goes.

Why did my rating change so much in my first games? A new rating starts at ±350, so each result moves it a lot until it settles.

I played a casual game / Chess960 / a game with a guest and nothing changed. Those games are never rated (see What counts for CSR).

I have never played Bullet, so why is my Bullet CSR not 1500? It starts from your other ratings (see Starting from your other ratings). It is marked “?” until a few Bullet games settle it.

Why did my rating not go down after a loss? It is at its floor: 400 points below your highest established rating, rounded down to a hundred (see Rating floor).

Why do I have five CSRs? Bullet and Correspondence chess are very different skills, so each speed has its own rating.

Why did my rating move more after a break? Your RD grew while you were away (see Inactivity), so the system lets your rating adjust faster.

My CSR is different from my practice rating. The practice rating comes from games against the computer on your device and is not CSR.

Why did my TSR not change after a Turbo Live event? The event was free (unrated), you cast fewer than 10 votes, or fewer than 2 players in it were rated.

Can I buy a better TSR? No. The entry fee only buys a place; top-ups (earned tokens only) keep you in the event but never change the agreement TSR uses.

Is my CSR my FIDE rating? No (see section 7).

Privacy policy · Terms of use · Safety & reporting · About & credits · Back to the game