TalkChess.com

[Marco Costalba]

Traditionally to the evaluation is given the meaning of measure of the position in terms of centipawns or, anyhow to measure how good is a given position assigning a number that more or less resembles the result of a chess position's analysis by a GM. This concept is IMHO completely misguiding.

Chess engines don't use evaluation to "analyze" a position, but to decide what moves to try first. If, for instance, we would have an evaluation function that returns random numbers but with the only constrain that given a position and a list of possible moves, the position obtained playing the strongest move has always the highest evaluation, well, this would be the perfect function from ELO point of view, but at the same time could be as well totally unusable for classical chess "analysis".