What Chess.com Accuracy Score Actually Means (And What It Misses)

If you’ve ever finished a chess.com game with an 85% accuracy score and still lost, you’ve already discovered the central problem with the metric: accuracy is not the same as good play, and it’s certainly not the same as understanding why you lose. Chess.com’s CAPS2 score is a useful number — but only if you know what it actually measures, and what it quietly leaves out. Here’s the honest breakdown of how the score is computed, why a single-game number is mostly noise, and what to look at instead when your goal is actually getting better.

How CAPS2 is calculated (the basic formula)

CAPS — Computer Aggregated Precision Score, currently in its second iteration as CAPS2 — is chess.com’s accuracy metric. The core idea is straightforward: for every move you played, the engine evaluates the position before and after. The difference between the engine’s preferred move and your move (the “evaluation delta,” measured in centipawns) tells the algorithm how much you gave up on that move. Sum that across the whole game, normalize, and you get a percentage between roughly 0 and 100.

A few details matter. CAPS2 doesn’t weight every move equally — moves in critical positions (where one side has a real choice between meaningfully different evaluations) count more than moves in dead-equal endgames where any reasonable choice is fine. The engine depth used is calibrated to balance speed and reliability — deeper than the in-browser real-time arrow, but not the kind of multi-hour analysis a correspondence player would use. The scoring also gets re-calibrated against rating bands, so a 95% accuracy doesn’t always represent the same wall-clock-perfect play across different time controls and opponent strengths.

The takeaway: the formula is reasonable, the math is honest, and CAPS2 is a defensible metric for what it’s designed to measure. The problem is that what it’s designed to measure isn’t quite what most players think it measures.

Why one-game accuracy is noise

Here’s the part most players miss: a single-game accuracy score has enormous variance. The same player, with the same skill level, will produce wildly different accuracy scores from one game to the next — not because they played differently, but because the position type changed.

Three structural reasons:

• Position complexity drives the score. A 30-move game where most positions had one obvious move scores higher than a 60-move game full of double-edged tactical decisions. Two players of identical strength, one playing a quiet Italian and the other a sharp Najdorf, will post different accuracies even if both played to their normal level.
• Opponent rating warps the result. When your opponent plays poorly, your “best moves” tend to be more obvious — capture the hanging piece, take the mate. Higher accuracy comes for free against weaker opposition. Against stronger players, the same skill level will produce a lower accuracy because the choices are harder.
• Decisive games skew toward extremes. A game you crushed in 22 moves often shows 90%+ for both sides because no one had to navigate a difficult middlegame. A long defensive grind looks worse on paper even if the defender played heroically.

Concrete example: Player A finishes with 85% accuracy and loses. Player B finishes with 75% and wins. What happened? Player A played a 50-move technical game where the few critical moments were all errors — high average, fatal in the moments that mattered. Player B played a tactical slugfest with several inaccuracies but found the one combination that decided the game. Average accuracy missed both stories. Critical-move accuracy — the score on the 3-5 turning points — beats average accuracy almost every time when you’re looking at a single game.

This is the core reason why obsessing over one game’s CAPS2 leads nowhere. The score is real, the inputs are real, but the signal-to-noise ratio at the per-game level is too low to drive any meaningful change in your training. If you want to use accuracy data productively, you have to zoom out.

What accuracy hides

Even when you average accuracy across many games, it still leaves out the things that actually drive losses. The score is a summary statistic — and like all summary statistics, it compresses away the structure underneath. Specifically, here’s what CAPS2 will never tell you:

• When in the game your mistakes happen. Two players with identical 78% accuracy can be radically different — one bleeds evaluation in the opening and recovers, the other plays cleanly until move 35 and collapses on the clock. They need opposite training plans. Average accuracy hides the timing.
• Whether mistakes cluster around specific patterns. If you keep losing to kingside sacrifices when the opposing bishop sits on h7, the accuracy score will register the blunder but won’t see the recurrence. Across 50 games the same motif might trip you up 8 times. CAPS2 sees 8 unrelated blunders. A coach sees a pattern.
• How time pressure correlates with accuracy. Your accuracy in the first 25 moves at 30 seconds per move is a different metric from your accuracy in moves 30+ at 5 seconds per move. The summary number averages them. The diagnostic insight requires separating them.
• What kind of player you are. An Aggressor and a Drifter can post the same accuracy in the same game while losing for completely different reasons. The number is identical; the prescription is opposite.

This is the archetype angle: average accuracy is a thermometer, not a diagnosis. It tells you the patient has a fever. It doesn’t tell you whether it’s a viral infection or a bacterial one. Treating the fever number — drilling random tactics until the score goes up — is exactly the kind of generic improvement work that fails 80% of adult learners.

Better metrics for improvement

If a single CAPS2 number is too noisy and an average across games is too compressed, what should you actually look at? Five metrics give you a more honest picture of your play, and most of them are derivable from the same chess.com data the accuracy score is built on:

1. Per-phase accuracy

Split your accuracy into opening (moves 1-15), middlegame (moves 16-30), and endgame (moves 31+). The shape of the breakdown tells you which phase to train. An 85/72/68 player needs middlegame and endgame work despite the strong opening number. A 60/82/85 player has an opening problem masked by middlegame skill. Chess.com’s Insights tab exposes some of this; the rest you can extract by manually flagging the phase of each error in 20 games.

2. Critical-move accuracy

Across 50 games, identify the 3-5 critical moments per game — the points where the evaluation could have shifted by half a pawn or more depending on your choice. Track your accuracy on just those moves. This is the metric coaches care about, and it’s typically 10-20 percentage points lower than your overall accuracy because critical moments are the hard ones. Your improvement target is to close that gap.

3. Blunder rate per phase

Count blunders (a chess.com Game Review label, or any move that drops evaluation by 2+ pawns) and stratify them by phase. Are 70% of your blunders in moves 30+? You’re Time-Pressured. Are 60% in moves 1-15? You’re Opening-Confused. The distribution names the archetype.

4. Time-pressure correlation

For each move, log the time spent. Then correlate move time with move quality. The healthy pattern: moves under 5 seconds are book moves with high accuracy, moves at 30-90 seconds are critical decisions with high accuracy, and moves at 5-15 seconds are routine moves with high accuracy. The unhealthy pattern: critical-decision accuracy collapses when time per move drops below 10 seconds. That’s a clock problem, not a chess problem, and the fix is opening prep and time allocation, not more tactics.

5. Archetype signal

The composite of the above four metrics — phase distribution of mistakes, critical-move performance, time correlation, and recurring pattern themes — maps onto one of five archetypes: Aggressor, Drifter, Time-Pressured, Opening-Confused, or Endgame-Soft. Each archetype has a different training prescription. The full archetypes guide walks through each profile in detail. The point is that the archetype is a diagnosis — actionable in a way a single accuracy number never is.

How to actually use accuracy in your improvement loop

None of the above means CAPS2 is useless. It just means you have to use it correctly. Here’s a simple framework that respects what the score is good for and routes around what it’s not:

1. Don’t react to single-game accuracy. A 92% on Tuesday and a 68% on Wednesday isn’t a story about your play. It’s a story about position complexity. Stop screenshotting one-game results.
2. Track the rolling 30-game average. A trend line of your last 30 games’ accuracy is meaningful — it smooths out per-game variance and reflects actual skill drift. If the line is climbing, your training is working. If it’s flat for three months, the training plan isn’t matching your actual weakness.
3. Use accuracy as a phase-distribution filter, not a verdict. When your overall average is good but your endgame phase score is bad, that’s the lesson. Train the gap, not the average.
4. Layer pattern analysis on top. Once a month, pull your worst 10 games and look for recurring themes — the recurring kingside sacrifice you miss, the recurring time scramble at move 28, the recurring inability to convert +2 endgames. The themes are what an accuracy score can’t see, and they’re what changes your rating.
5. Re-diagnose when your rating moves. Your archetype shifts as you improve. The Time-Pressured 1200 might be an Opening-Confused 1400 six months later. Re-run the diagnosis when your rating drifts 100+ points so the training plan stays calibrated.

If you want to do this manually, the four-step coach-style review in our guide on analyzing chess.com games like a coach walks through the workflow on a per-game basis — write your reasoning before reading the engine, classify the loss, find the recurring theme, build one drill from the pattern. It’s the same logic the better accuracy metrics aim at, just done by hand.

If you want the 100-game pattern view in 60 seconds instead of 6 hours, that’s exactly what we built MyChessPlan for. The free report classifies your last 100 chess.com rated games into one of 5 archetypes, breaks down your accuracy by phase, flags your time-pressure correlation, and produces a 7-day starter plan calibrated to your rating band. It’s not a replacement for the analysis habit — it’s a way to skip the bookkeeping so you can spend your time on the drill, not the spreadsheet. And if you’ve felt stuck on the same losing patterns no matter how many games you play, our deep-dive on why you keep losing the same way covers the cognitive science underneath.