How do you know if you are improving at chess?

A rating is a comparison with the people you are playing now. One night, or one number from years ago, is a weak way to judge your own play.

Lichess uses Glicko-2, and Chess.com's help center describes the Glicko system. Both change your rating more when they are less confident in it, so big early swings, or swings after a break, are expected. They are not proof you got worse. What "good" means inside one pool is what is a good chess rating. Why the two sites differ is Chess.com vs Lichess rating.

On Lichess, ratings are calculated with the Glicko-2 method developed by Mark Glickman. The Lichess FAQ says a new rating starts at 1500 ± 1000. The 1500 is the rating, and the 1000 is the confidence interval. Lichess says the system is 95% sure that rating is somewhere between 500 and 2500, and that it is incredibly uncertain. Because of that, "when a player is just starting out, their rating will change very dramatically, potentially several hundred points at a time." After some games against established players, "the confidence interval will narrow, and the amount of points gained/lost after each game will decrease."

Chess.com's help article does not state a starting rating, and it does not state a swing size. Lichess's rating-systems page says new Chess.com players now choose their initial rating. Chess.com's own article says that if you are new and have not played any games, the system is unsure of your real rating, "so your rating will fluctuate significantly during your first few games." A new Chess.com rating can move a lot. That article does not call the move "several hundred points." That phrase is Lichess's, about a new Lichess rating.

What the Lichess question mark means

The Lichess FAQ says: "The question mark means the rating is provisional." One reason is that the player has not yet finished enough rated games against opponents of similar strength in that category. Another is inactivity: "The player hasn't played enough recent games. Depending on the number of games you've played, it might take around a year of inactivity for your rating to become provisional again."

Lichess then says: "Concretely, it means that the Glicko-2 deviation is greater than 110. The deviation is the level of confidence the system has in the rating. The lower the deviation, the more stable is a rating." Read a rating with a question mark as unsettled. Wait until the mark is gone before you treat the number as a trend.

A break makes the next games count more

Lichess words the inactivity effect this way: "as time passes, the confidence interval will increase. This allows you to gain/lose points more rapidly to match any changes in your skill level over that time."

Chess.com's help center says the same idea with rating deviation. "If you haven't played in a while, your RD will increase, causing your rating to change more dramatically after your next game. This happens because we can't determine whether you've been practicing elsewhere or haven't played at all." High confidence means smaller changes. Low confidence can mean bigger changes. As players reach higher levels on Chess.com, "the rating changes after each game become smaller."

The size of a Chess.com change also depends on the gap between the two ratings, and on confidence in the opponent's rating as well as yours. A win against a much higher rating usually moves you more than a win against a similar rating. A loss to a much lower rating usually moves you more than a loss to a higher rating. Neither site publishes a "normal" swing for an established rating. A move of 100 points is not, by itself, a verdict.

A rating compares you with the people in that pool. Lichess's FAQ says it is best not to think of ratings as absolute numbers. "It's best to think of ratings as 'relative' figures." Inside one pool, the gaps help you estimate who wins, draws, or loses, and how often. "Saying 'I have X rating' means nothing unless there are other players to compare that rating to."

Lichess's rating-systems page says ratings cannot be compared directly from one server to another. "They only serve to compare the skill of players within the same rating pool." In our view, the same caution applies across years on one site, because the pool of active players changes. Where a number sits among those players can change too. A phrase like "better than most players" can drift even when your own games look steady, because it describes the active pool, not a fixed ladder.

Inflation, in the ordinary sense, would mean the numbers in a pool rise with no matching rise in skill. Deflation would mean they fall with no matching drop in skill. This is a concept, not a measurement. This guide does not claim that Lichess or Chess.com is inflating or deflating. A rating from two years ago and a rating today are two comparisons, often with two different crowds. They are a weak before-and-after of your skill.

One time control, over two or three months

Pick one time control and stay with it. On Lichess, wait until the question mark is gone, which means the deviation is no longer above 110. On Chess.com, give the first games, and the games right after a long break, time to settle, because low confidence makes those changes larger.

Then write the rating about once a month for two or three months. Two or three months is a practical window, not a law. Read the direction across those notes. One evening is noise. A drop inside that window is still not proof of lost skill. Habits, tired sessions, and a new opening can move the number. The checks for that are in why am I getting worse at chess.

Mistakes you can count

Review games you played, and count the mistakes you can name. A practical count is blunders per game in the games you actually reviewed, plus the tactical miss that keeps coming back, such as a hanging piece. Replay without an engine first. That method is how to analyze your own chess games. The habit behind a repeated blunder is how to stop blundering.

Compare a month with the next month. A lower count in similar games is a stronger signal than a rating that bounced for a weekend.

Average centipawn loss

Lichess defines average centipawn loss in its FAQ. A centipawn is 1/100th of a pawn, so 100 centipawns equal 1 pawn. "The top computer move will lose zero centipawns, but lesser moves will result in a deterioration of the position, measured in centipawns." "The fewer centipawns one loses per move, the stronger the play." Lichess computer analysis is powered by Stockfish.

That number depends on the opponent and on the position. A quiet game and a sharp game do not produce the same loss even when your care is similar. Compare your own trend across similar games. One spectacular game, or one game against a much stronger opponent, is a weak month.

The phase, or the clock, that keeps failing

After several reviews, group the mistakes by phase, and note losses on time. The label that returns is the one to train. How to rank those labels is how to find your chess weaknesses. A single loss in an ending, or a single flag, is a story. The same label across a month is a trend.

The same puzzle set, or the same engine level

Optionally, repeat one puzzle set, or one fixed engine level, about once a month, and write the result next to the date. Keep the set or the level the same, or the comparison moves.

The limit is built in. You can remember a puzzle. An engine does not rush, bluff, or tilt the way a person does. A better puzzle score is a hint that those positions got easier. It is not a rating, and it does not replace the games you play against people.

Once a month is a practical rhythm. It is guidance, not a rule. Checking every night usually follows the swing instead of the trend.

1. Write the date, the site, and one time control. Note whether the Lichess rating still has a question mark, or whether you are in the first games, or just back after a break, on Chess.com.
2. Write the rating. Compare it with the notes from the last two or three months, not with a number from years ago.
3. From the games you reviewed, write blunders per game and the tactical miss that showed up most often.
4. If you have an analysis, write the average centipawn loss, and remember it depends on the opponent and the position.
5. Name the phase, or the clock problem, that repeated. Pick that one thing to practice before the next check.
6. Optional: write the result of the same puzzle set or the same engine level, with the limit above in mind.

Run a Chess Fingerprint on the games you have now, then run a deep scan again later, about a month if that is your rhythm, and compare the scans. The fingerprint suggests a focus from its top leak; treat it as a signal, not a verdict.

The report ranks up to three leaks and shows the first one as the next action. The same leak list goes into the weekly plan. A leak is added when its rule fires. Examples: at least 15% of losses on time with clocks in at least 10 games; at least 35% of moves after move 10 played in under 5 seconds, across at least 200 moves; an opening score of 40% or less over at least 4 games. When shorter games score at least 10 points higher and there are at least 10 longer games, one card is titled "Long games slip away — your endgame is the leak," and its fix is endgame study. Another leak can still rank first. When none of the leak rules fire, the report shows a default card titled "No single dominant leak — sharpen blunder control."

A deep scan is separate. It reports average centipawns lost per move by phase, plus blunder and mistake counts. A blunder is an evaluation drop of at least 200 centipawns, and a mistake is at least 100 and under 200. Positions already decided on both sides of the move are not counted as blunders. The phase with the highest average is the one the scan marks as needing the most work. It also groups blunders by type, including a hung piece or a missed tactic.

Saved scans keep those phase averages and a blunder rate per 100 moves. The browser keeps up to 30 scans for that username, and at most one per hour. A signed-in account can sync them. When two scans are saved, Progress compares the latest with the previous one. That is a comparison of scans you ran, not a rating history. The rating chart on the report is the path through the recent games just read, in one time control. It is not a record of your rating from two years ago.