Why ACPL is Misleading: The Mathematical Genius of MQ-Chess (Loss Index)

# Why ACPL is Misleading: The Mathematical Genius of MQ-Chess For decades, chess players relied on **Average Centipawn Loss (ACPL)** to evaluate the accuracy of their games. However, any grandmaster or software engineer knows that ACPL is deeply flawed. ## The Two Core Flaws of ACPL ### 1. The Superfluous Advantage Problem Imagine you are playing White and reach a position with an evaluation of **+9.00** (up a queen and a rook). You play a move that drops the eval to **+8.00**. - Under ACPL, you are penalized **100 centipawns of loss**. - In reality, your win probability remains **100%**. The move had **zero practical impact** on the outcome of the game. Conversely, consider a sharp equal position at **+0.10**. You make an inaccuracy that drops the eval to **-0.90** (dropping from equal to completely losing). - Under ACPL, you are also penalized **100 centipawns of loss**. - In reality, your win probability collapsed from **50% to 15%**. ### 2. Ignoring Decision Complexity ($N_{\text{legal}}$) In a forced endgame position with only **2 legal moves**, choosing the correct move requires far less cognitive calculation than a midgame position with **45 legal moves**. ACPL treats both decisions identically. ## The MQ-Chess Solution: Win Chance & Complexity Weighting MQ-Chess introduces two mathematical transformations: 1. **Sigmoid Win Chance Mapping**: $$\sigma(cp) = \frac{2}{1 + e^{-0.003624 \cdot cp}} - 1$$ 2. **Decision Complexity Weighting**: $$w = \ln(N_{\text{legal}} + 1)$$ 3. **Loss Index Formula**: $$\text{Loss Index \%} = \frac{\sum |\sigma(\text{eval}_{\text{best}}) - \sigma(\text{eval}_{\text{played}})| \cdot \ln(N_{\text{legal}} + 1)}{\sum \ln(N_{\text{legal}} + 1)} \times 100\%$$ By measuring **Win-Chance Delta** weighted by **decision options**, MQ-Chess provides an objective, un-cheatable measurement of true decision precision.