Run any game review and your moves come back wearing labels: Best, Excellent, Inaccuracy, Mistake, Blunder. The labels feel authoritative, a red double question mark stings like a teacher's pen, but almost nobody explains where the lines between them actually sit. Once you know, the labels become far more useful and considerably less scary.
It's all about win probability
Under the hood, an engine evaluates the position before and after your move. The modern approach converts those evaluations into a winning probability and measures how much your move dropped it. A move that keeps your chances where they were is fine, whatever it looks like. Roughly speaking: shave off a couple of percent and it's an inaccuracy; give away a chunk and it's a mistake; throw away ten or more points of win probability in one go and you've earned the blunder tag. Rookie's own grades (Wobble, Whoops, Facepalm) draw their lines the same way.
The win-probability framing explains something raw engine numbers get wrong: context. Losing 150 centipawns in a dead-equal position might swing the game from draw to loss, a genuine blunder. Losing the same 150 centipawns when you're up a queen changes your winning chances by almost nothing, and a good grader barely dings it. The same engine drop is graded differently depending on how alive the game still was, which is exactly how it should work.
Which labels deserve your attention
Not all of them equally. Inaccuracies are mostly noise for improving players; even strong club players produce them constantly, and chasing them is premature optimization. The blunders and mistakes are the game-deciders, and they cluster into patterns: a specific tactic you keep missing, a type of position where you reliably drift. One genuine blunder pattern, found and fixed, is worth more than fifty polished inaccuracies.
Also worth knowing: "Best" doesn't mean you had to find the engine's top move. Playing the second-best move in a position where three moves all hold the balance is perfect, practical chess. The label system rewards not losing your footing, not machine imitation.
Using the labels without being used by them
Treat a review's labels as a map of where to look, not a verdict on your worth. Open the two or three worst-labeled moments, work out what you missed and why, and give the rest of the labels no more attention than they've earned. And when the same label keeps landing on the same kind of move across games, that's not bad luck. That's your improvement plan writing itself.