Chess DNAWhich phase of chess decides your rating?
From under-600 to 1800+, opening accuracy climbs 13 points. The middlegame adds 4. Endgame accuracy doesn't move at all — and beginners post the best endgame number in the data. That last one is an illusion worth understanding.
164 players, bucketed into six rating bands of 23–34 players each · accuracy computed on the player's own moves only.
Three phases, one ladder
Average accuracy per phase, by rating band. The endgame line sits flat at the top for everyone. The opening line is the one doing the climbing — 70.3% → 83.7% across 1,200+ Elo of skill.
If phase accuracy measured phase skill, this chart would say beginners are elite endgame players who can't handle openings. One of those two claims is real. The opening climb is the honest one: openings are the only phase where every player faces roughly the same problem — develop, castle, don't drop material — so accuracy there tracks skill cleanly. Opening inaccuracies fall from 4.3 per game under 600 to 2.9 at 1800+, and in our companion analysis of what each Elo level struggles with, opening mistakes were the single most common tagged theme at every band.
The middlegame line is the strange one — it barely climbs, and it isn't even monotonic. The 1500–1799 band posted 75.8%, lower than the under-600 band's 77.8%. That's partly small-sample noise, and partly something real: stronger opponents create harder middlegames. Your positions get more difficult at roughly the rate you get better at handling them, and the average barely moves.
The 93% endgame is an illusion
Players rated under 600 post 93.1% endgame accuracy — the highest phase number in the entire dataset, above anything the 1800+ band produces in any phase. They are not endgame wizards. By the time a sub-600 game reaches an endgame, somebody has usually already lost it: a queen was hung on move 19, and the rest is mop-up. In a decided position, nearly any reasonable move keeps the evaluation — the engine scores this as near-perfect play. Endgame accuracy mostly measures how decided the game already was, not endgame skill.
Two corroborating signals from the same dataset. First, in blitz the illusion runs backwards: endgame accuracy falls 3.7 points (92.0% → 88.4%) from under-600 to 1500+, because stronger players reach genuinely contested endgames — close positions where real mistakes are possible. Second, when we count engine-tagged endgame-technique mistakes per game instead of accuracy, the number rises with rating: 1.28 for beginners, 1.98 at 1800+. Stronger players don't get worse at endgames — they finally have endgames to get wrong. If your improvement plan skips endgames because your accuracy report says 92%, the report is lying to you; here's how to actually improve your endgame.
Where the climb actually happens
Accuracy points gained from under-600 to 1800+, per phase. The opening carries +13.4 of the climb; the endgame contributes nothing.
This shape survives every robustness check we threw at it. Restricting to rapid games only — 68 players, 3,540 games, opponents within ±250 Elo — the climb is +16.0 opening, +5.1 middlegame, +0.2 endgame. In blitz, the opening still climbs +8.4 while the endgame drifts down. Different subsets, different time controls, same answer: ratings diverge early in the game. The phase where players separate isn't the one with the highest error stakes — it's the one where skill expresses itself most reliably, move after move, game after game.
What this means for your study time
Below ~1200, the opening gap is where your rating lives — but "study openings" doesn't mean memorize lines. The sub-600 to 1200 climb is almost entirely principled play: develop fast, castle early, stop shedding pawns and pieces to two-move tactics in the first fifteen moves. That's opening discipline, not opening theory. A small, boring repertoire you actually understand beats a deep one you half-remember.
Don't let a 92% endgame accuracy talk you out of endgame study. The number is inflated for everyone, at every level, because decided positions score as free accuracy. The better question is how often technique decides your close games — and that rises with rating. From about 1200 up, more of your games reach genuinely equal endgames, which is exactly when technique starts converting directly into rating.
And the middlegame? Its accuracy average hides where games are actually decided. Middlegame improvement mostly shows up as fewer catastrophic single moves, not a prettier average — the average size of a player's worst move shrinks from 723 centipawns under 600 to 492 at 1800+. We unpack that in how chess games are actually lost, and which mistake types drive it in which mistakes go away as you improve.
Opening accuracy points from under-600 to 1800+ (70.3% → 83.7%) — the steepest phase climb.
Endgame accuracy of players rated under 600 — the highest phase number anywhere in our data.
Blitz endgame accuracy change from under-600 to 1500+. Stronger players reach harder endgames.
Endgame-technique mistakes per game, beginners → 1800+. Real endgame skill is invisible to phase accuracy.
Questions people actually ask
Which phase of chess should I study to gain rating?
The phase that separates rating bands most sharply is the opening: 70.3% accuracy under 600, 83.7% at 1800+ — a 13.4-point ladder, versus 3.9 in the middlegame and zero in the endgame. But below ~1200 that gap is principled play (develop, castle, stop dropping material early), not memorized theory. Openings are where weak play gets punished most reliably, so improvement shows up there first.
Why is endgame accuracy so high even for beginners?
Because their endgames are usually already decided. When one side is up a queen, nearly any move keeps the evaluation, and the engine scores that as near-perfect play. Sub-600 players post 93.1% endgame accuracy precisely because someone blundered long before move 40. The stat measures how decided the game was, not endgame skill.
Does this mean endgames don't matter?
No — it means accuracy can't see endgame skill. Count engine-tagged endgame-technique mistakes per game instead and the number rises from 1.28 (beginners) to 1.98 (1800+): stronger players finally reach contested endgames worth getting wrong. As you climb, technique in equal positions converts directly into rating — endgame study pays more, not less.
Why doesn't middlegame accuracy improve with rating?
It gains just 3.9 points, and not monotonically — 1500–1799 players posted 75.8%, below the under-600 band's 77.8%. Middlegames get harder as you climb: stronger opponents create harder problems, and tension survives longer. Improvement shows up as fewer catastrophes — the average worst move shrinks from 723 to 492 centipawns — not a prettier average.
How was phase accuracy measured?
164 Chess DNA players, full game history, Stockfish 17 at depth 18. Moves are split into phases by the app's phase detector, and accuracy is computed on the player's own moves only. The rapid-only subset (68 players, 3,540 games) reproduces the shape: opening +16.0, middlegame +5.1, endgame +0.2. Full gates in the method note below.
Want to see your own phase profile? Chess DNA analyzes your games with Stockfish and shows your accuracy by phase, your recurring mistakes and where your rating actually leaks — the same pipeline that produced this report.
Method. 164 Chess DNA players with full-history Stockfish 17 depth-18 analysis, bucketed by average game rating into six bands: <600 (n=34), 600–899 (29), 900–1199 (25), 1200–1499 (25), 1500–1799 (28), 1800+ (23). Phase boundaries come from the app's phase detector (development and material based); accuracy is computed on the player's own moves only, so opponents' play never touches the number. Endgame-technique mistake counts come from the same engine-tagged theme pipeline. Robustness: the rapid-only subset of our time-control dataset (68 players, 3,540 games, ≥15 analyzed games per player, opponents within ±250 Elo) gives opening +16.0 / middlegame +5.1 / endgame +0.2; blitz gives opening +8.4 with endgame declining −3.7. Phase accuracy is a signal about where ratings diverge — not a per-phase skill grade; decided positions inflate endgame accuracy at every level. This is part of a series built on the same dataset as What Each Elo Level Struggles With.