Learn · Engines

Reading an engine like a pro

What the numbers mean, why depth matters, how accuracy is worked out, and when you should ignore the computer completely.

8 min readUpdated October 2026

What the evaluation means

The engine's score is measured in pawns from White's point of view. Plus is good for White, minus is good for Black.

ScoreWhat it usually means
0.00 to ±0.30Equal. Either side can win with better play.
±0.30 to ±0.90A small edge. Pleasant to play, not a winning advantage.
±0.90 to ±2.00A clear advantage. Often winning between strong players.
±2.00 and aboveWinning. Usually equal to an extra piece or a crushing attack.
+M5 or −M5Forced checkmate in five moves for White (plus) or Black (minus).

These numbers describe perfect play from here on. A +1.5 position can still be very hard for a human to win, and a −3 position can be easy to save if the only way to win is ten precise moves long.

Depth and thinking time

Depth is how many moves (strictly, half-moves) ahead the engine has searched along its main line. Deeper is more reliable. In quiet positions depth 18 to 20 is already very strong; in sharp tactical positions, give it longer. If the best move keeps changing as depth grows, the position is complicated, and that is useful information too.

Use more than one line

The analyser shows the top three moves. Compare their scores:

  • If all three are close, the position is flexible. Pick the move you understand best.
  • If the best move is far ahead of the second, there is only one good move. Work out why before moving on; that is where the lesson is.

How accuracy is calculated

The evaluation is first turned into a winning chance for each side. A position at +1.00 gives White roughly a two-in-three chance of winning between strong players. Each move is then judged by how much winning chance it gave away. Losing 10 points of winning chance is an inaccuracy, 20 a mistake and 30 a blunder. Accuracy is an average over the game of how close each move came to the best.

That is why a move that drops a pawn in a totally winning position barely counts, while the same pawn in an equal position is a mistake: it is the effect on the result that matters.

Local engine, cloud and tablebase

  • Stockfish 18 runs on your own device using WebAssembly. It is the lite single-thread build, which is light enough for phones and still plays at super-grandmaster strength.
  • Lichess cloud stores deep analysis for millions of positions. For well-known positions you get a depth 40+ answer instantly.
  • The tablebase is a different kind of tool: it does not search, it looks up the perfect result for every position with seven or fewer pieces.

When to ignore the engine

  • In the opening. A move rated +0.2 instead of +0.3 does not matter. Play what you understand.
  • When the best line is inhuman. If it needs ten only-moves, choose a simpler line that keeps a good position.
  • During a game. Never. Using an engine against a person is cheating on every site and in every federation.

A good analysis routine

  1. After a game, go through it without the engine first. Mark where you were unsure.
  2. Run the game review.
  3. For each mistake, find the better move yourself before you look. Then compare.
  4. Write one lesson per game. Patterns repeat; so will the lessons.