Evaluation function

An evaluation function gives a numerical estimate of how favorable a position is for one side. A chess engine uses it when finite search cannot reach the actual end of every possible game.

The evaluation is called static because it scores the position at that point without starting another full search there. Static does not mean tactically quiet.

Information it may consider

• Material, meaning which pieces each side still has.
• King safety, piece placement, and mobility, meaning how freely the pieces can make useful moves.
• Pawn structure, meaning how the pawns are placed and support or weaken one another, along with other positional patterns.
• In some engines, a neural network, a mathematical model trained on many examples to combine these signals.

The sign and scale require context. A positive number may favor White or the side to move, depending on the program. A unit resembling one pawn also does not represent a universal winning probability.

The function does not choose the final move by itself. Search compares many future positions and carries their values back toward the starting position through methods such as alpha-beta pruning. A quick evaluation of the current board can therefore differ from the result after possible move sequences have been searched.