More games will not raise your rating on their own

The short answer

Playing more games raises your rating only up to the point where you stop meeting new problems. After that, volume mostly rehearses the mistakes you already make. A player who plays 300 blitz games a month and never looks at one of them is practising their current error set with great diligence.

The honest version is that games are the input, not the training. They generate the raw material — positions you handled badly. Something has to turn that material into a change in what you play next time. That something can be an engine, a coach, a notebook, or ten minutes of thinking on the train. It cannot be nothing.

What your mistakes actually look like

Across 232,038 engine-classified moves from 7,329 analysed games on Mated, 3.8% of moves played are outright blunders and another 4.2% are mistakes. That is roughly one real error every 13 moves, before you count inaccuracies, which run at 18.4%. In a 40-move game you are making three or four moves that meaningfully change the evaluation.

Those errors are not spread evenly. In the same 232,038 moves, 9.1% of middlegame moves are a mistake or worse, against 2.3% in the endgame. Part of that is that endgames have fewer plausible candidate moves. But it also means the phase where most players spend their study time — openings — is not where the damage is, and the phase they avoid studying is not costing them as much as they assume.

One more number that changes how you should read your own reports. In that same set, the average move gives away 93 centipawns against the engine's best, but the median is far lower. A small number of very bad moves carry most of the damage. You are not losing games by drifting. You are losing them on three moves, and the other 37 are fine.

That matters for volume. If your losses come from a handful of identifiable moments, more games only helps if you are finding those moments. Playing another fifty games adds another 150 of them to a pile you are not reading.

When extra games make you worse

Some of these are obvious and you still do them.

On the tilt point specifically: nobody has measured the size of the effect well at club level, as far as we know. But the mechanism is not mysterious. You play faster, you stop checking your opponent's last move, and you lose to the same tactic three times in an hour.

• Playing on tilt. The game after a bad loss is played by a worse player than you. So is the one after that.
• Dropping time control to fit more in. Bullet is a fine game. It does not train the calculation habits that decide your 10+0 games, and if bullet is 80% of your volume your slower results will reflect that.
• Farming weaker opponents. Winning on material against 400-rated accounts does not produce positions you had to solve.
• Repeating the same opening into the same middlegame structure with no idea why you keep losing it. That is not repetition building skill, it is repetition building a rut.
• Playing when tired. Your blunder rate is not constant across the day and you know roughly when yours goes up.
• Playing instead of reviewing, because reviewing feels like admin and playing feels like chess.

A concrete comparison

Take two weeks of the same 1200-rated player. Week one: 60 blitz games, no review, rating ends where it started plus or minus 30 points of noise. Week two: 20 games, and after each loss they open the game and find the single move where the evaluation swung most, then write down what they missed in one sentence. Not a plan, not a study programme. One sentence.

By the end of week two they have 12 or so sentences. Half of them say something close to the same thing — "I moved a piece and left the knight on f6 hanging to the bishop", "I took on d4 without checking the recapture". That repetition is the finding. You cannot get it from playing. You can only get it from looking at several games at once.

This is the part that is tedious enough that people do not do it, which is why Mated does the reading step for you: it pulls your recent games from the Chess.com and Lichess APIs, runs Stockfish over every position in the browser, and turns the pattern in your own errors into a fifteen-minute session. But the manual version works. The thing that does not work is neither.

How many games is right

There is no published number and anyone who gives you one is guessing. What can be said is structural: your games-per-week should be low enough that you can review a meaningful share of them, and high enough that you are playing at least a few times a week so the review has something recent to attach to.

For most improving adults that lands somewhere between five and twenty serious games a week, with a longer time control for at least some of them. If you play 100 and review none, cutting to 30 and reviewing ten is very likely a better week. If you play three and review all three, you probably need more games, not more analysis — three games is too small a sample to show you a pattern.

One caveat worth being honest about: rating also moves for reasons unrelated to skill. Pool inflation, matchmaking variance, and a 40-point swing from six games are all real. Do not judge a change in habits on two weeks of rating.

The same trap exists in puzzles

Volume without difficulty fails the same way in tactics. Across 2,107 rated puzzle attempts on Mated, 94.3% are solved first time. That is a pleasant number and a bad one. A puzzle set pitched at your actual edge should be failing you regularly — somewhere near a third of the time is a reasonable target, though that figure is a judgement call rather than a measured one.

If you are clearing puzzle after puzzle on sight, you are drilling recognition of patterns you already own. The sensation is productivity. The content is revision. Push the rating up until you start getting things wrong, then stay there.

Rating is the slowest instrument you own

If you want to know whether a change is working, rating will tell you in about three months. Your error rate will tell you in about three weeks. Count blunders per game, or count the games where you were winning and did not convert. Those numbers move before the rating does, and they move for reasons you can point at.

There is also a category of game that teaches you more than the rest, and it is not your wins or your crushing losses. It is the games you nearly lost and did not — the ones where you were down a piece at move 20 and swindled it back. You will not review those, because the result was fine. Review them anyway. The mistake is still in there, and the only reason you did not pay for it is that your opponent made a bigger one.