ACPL — Average Centipawn Loss

ACPL

Definition

ACPL stands for Average Centipawn Loss. A centipawn is one one-hundredth of a pawn (100 centipawns = 1 pawn). ACPL measures, on average, how far a player’s moves deviate from the engine’s top choice throughout a game. Lower ACPL means closer-to-optimal play; higher ACPL indicates larger mistakes or a generally lower quality of moves.

How it is used in chess

• Post-game quality check: Players review ACPL to gauge overall precision without reading every engine line.
• Comparing performances: ACPL lets you compare your accuracy across games, openings, or time controls.
• Training focus: Spikes in ACPL often highlight phases (opening, middlegame, endgame) where you’re hemorrhaging value.
• Broadcasts and commentary: Commentators use ACPL or related metrics to summarize how cleanly players performed in elite events.
• Fair-play analytics (with caution): Persistently, suspiciously low ACPL across many games may raise flags, but ACPL alone is not proof of anything.

How it’s calculated

For each of your moves, an engine evaluates the position and computes the difference, in centipawns, between the evaluation after your move and the evaluation after the engine’s best move. These differences (your “losses”) are averaged.

• Centipawn loss on a move = Eval(best move) − Eval(your move), measured in centipawns from your perspective.
• ACPL = (Sum of your centipawn losses over considered moves) ÷ (Number of considered moves).
• Platforms differ: Some exclude book moves, forced moves, or final checkmates; some round or cap extreme values; some analyze at different engine depths.

Interpreting the numbers

• Single-digit ACPL: Elite, often seen in strong classical games or very clean wins.
• 10–25 ACPL: Strong, typical for advanced play, especially with adequate time.
• 25–50 ACPL: Common at club level or in faster time controls where errors are more frequent.
• 50+ ACPL: Indicates substantial inaccuracies, time trouble, tactical oversights, or blunders.

Context matters: fast games, sharp openings, time trouble, and low engine depth typically push ACPL higher. A completely winning position can also inflate ACPL because many “merely good” moves score much worse than the engine’s fastest conversion.

Examples

Example 1: A short, wild sequence with clear mistakes. Two early blunders balloon ACPL.

Try the mini-trap below and notice how a single bad move can dominate your average:

• After 5...Nd4??, White wins material with 6. Nxe5. That move alone can add hundreds of centipawns to Black’s loss.
• Another blunder like 7...Qg5?? compounds the problem, sending Black’s ACPL soaring even if other moves were reasonable.

Example 2: A calm opening with few errors often yields low ACPL if both sides play natural, solid moves:

• These are mainline Ruy Lopez moves. If both sides follow well-known plans, the ACPL for each side tends to stay low.

Example 3: A toy calculation to illustrate the average. Suppose your centipawn losses over five moves are 12, 0, 18, 50, and 20. Then your ACPL = (12 + 0 + 18 + 50 + 20) / 5 = 20.

Strategic and historical significance

As engines became standard in analysis, ACPL emerged as a simple, comparable indicator of precision. It complements qualitative assessments: a player might choose a practical plan that’s –30 centipawns from the engine’s dream line yet easier to play for humans. Elite events often feature single-digit ACPL stretches, reflecting near-flawless technique under pressure. Commentators during matches like Kasparov vs. Deep Blue (1997) and modern World Championship contests have used centipawn swings and averages to highlight how closely humans match engine choices in critical moments.

Limitations and pitfalls

• Engine settings: Depth, multi-PV, and time spent affect evaluations and thus ACPL.
• Forced vs. free choice: In forced lines, there’s little room to lose centipawns; in complex positions, small human deviations can look “big” numerically.
• Winning conversions: When you’re already +5, many human moves are fine, yet the engine may prefer a faster mate, inflating ACPL.
• Non-uniform thresholds: Definitions of inaccuracy/mistake/blunder and which moves “count” differ across platforms.
• Not a style metric: Positional, prophylactic, or practical decisions may score worse than flashy engine tactics but still be excellent over-the-board choices.
• Not stand-alone evidence: ACPL by itself isn’t conclusive for fair-play judgments.

Related terms

• Centipawn: The unit of value used by engines (100 = one pawn).
• Accuracy: A percentage-based quality metric some platforms display alongside ACPL.
• Blunder / Mistake / Inaccuracy: Labels tied to centipawn thresholds that differ by site.
• Engine: Post-game evaluation using programs like Stockfish.
• Evaluation: The engine’s numeric assessment of who stands better and by how much.

Interesting facts and anecdotes

• World Championship games frequently feature long stretches where both players’ ACPL stays in the single digits, underscoring extraordinary precision under stress.
• In blitz and bullet, even top grandmasters can post ACPL figures far higher than in classical games due to time pressure.
• Some platforms report separate ACPL for White and Black, letting you spot color-based trends in your play.
• Sharp openings can lead to higher ACPL because one slip often costs a lot; solid systems tend to keep ACPL modest.

Practical tips to improve your ACPL

• Slow blunder-check: Before each move, ask “What are my opponent’s forcing replies?” Catching tactics cuts huge centipawn losses.
• Endgame basics: Knowing key endings reduces conversion errors the engine penalizes heavily.
• Opening hygiene: Learn main ideas to avoid early tactical landmines that spike ACPL.
• Time management: Preserve time for critical positions where a single slip can cost hundreds of centipawns.
• Targeted review: Filter your games for phases with the largest ACPL spikes and train those patterns specifically.