Accuracy is so High , BUT

I was in big disadvance in some period on analysis graph. So how was accuracy in all game phases near 90%, but such loss in evaluation, centi is not bad too , around 33.
So is game in common counted as good, if I play quite good, independed of figure loss, or there is something else makes game counted as good, independed of such disadvance periods?

[Event "rated blitz game"] [Site "https://lichess.org/3AU1LI8k"] [Date "2026.08.22"] [White "K3W719"] [Black "aVague"] [Result "0-1"] [Variant "Standard"] [TimeControl "180+2"] [Annotator "lichess.org"]
37 Qg7 Nf5?? (0.55 --> 5.61) Blunder.
49. Qxd5+?? (6.10 --> -8.28) Blunder.
Blunders tend to wash out the effects of other moves. It is probably not-so-great if things like that are going on in your games, but perhaps they are to be expected at 3+2.

whether a game is good or not is a difficult question. i suggest the following rule: if you remember its details for a month, it is good; if you forget, it is not.

There are several reasons why White's average centipawn loss is low.

1. Opening style: the opening of this game is closed (1.d4) and particularly cramped for both sides (two sorts of Colle systems facing each other); consequently, there are generally less inaccuracies and centipawn loss than in an open opening (1.e4) where a lot of tactics can be overlooked.

2. The first 20 moves are entirely sound, with the exception of two inaccuracies on move 18 (since Black had clearly planned to play f5, White's move 18.Bg4 wastes two tempi; and since Black could have played f5 as planned, 18...Bxg4 is also bad). That is all.

3. Later on, around move 22, the obvious plan for both sides was to play against the backward pawns, but at first glance, this appeared practically impossible; in my humble opinion, this is why there is no significant centipawn loss.

4. Next, it is a drawn endgame: Stockfish consistently evaluates such positions at 0.0. Therefore, there cannot be any centipawn loss at this stage either.

5. After the blunder on move 37, Black is clearly lost (a position evaluated as a 5-pawn deficit). When a position is so unbalanced, the algorithm will always find all moves perfectly accurate... even if there are make mistakes around one pawn down! The same on the final blunder, with an evaluation of the position now entirely in Black's favor.

6. Eventually, when you divide the size of the only big blunder on each side by 61 (the number of moves), they are completely diluted in the whole. I even think there is a cap in the accuracy calculation algorithm, and that an error exceeding 2 pawns is always capped at 2 pawns, or something like that.

@Goran_Pastek said ^

There are several reasons why White's average centipawn loss is low.

1. Opening style: the opening of this game is closed (1.d4) and particularly cramped for both sides (two sorts of Colle systems facing each other); consequently, there are generally less inaccuracies and centipawn loss than in an open opening (1.e4) where a lot of tactics can be overlooked.

2. The first 20 moves are entirely so...

Much Thx, Idea of 5 is interesting