Engines convert their assessment of a position into a single number measured in pawns (or centipawns), where positive favors White and negative favors Black. The scale lets you compare moves precisely: a move that drops from +0.50 to -0.20 has handed the opponent the better game even if no material changed hands.
Centipawn loss is also the basis of game-accuracy metrics — averaging how far each of your moves fell short of the engine's best reveals how cleanly you played. Treat the numbers as guidance rather than gospel, since small evaluations often reflect imbalances that are hard for humans to convert.