Incorrect statistics in replay

[thiccyoshi4568]

Type

Game behaviour

Bug description

Got a score with 95.00% and watched the replay. The accuracy after the replay somehow increased to 95.13%. Looking at the screen recording, it seems that the judgement counts changed. I'm so sorry for the watermarks and bad video quality by the way.

Screenshots or videos

clideo_editor_5935fceadf744c6eac35706888ce3275.mp4

Version

2022.810.2-lazer

Logs

performance-audio.log
performance-draw.log
performance-input.log
performance-update.log
runtime.log
updater.log
database.log
network.log
performance.log

[AuracleTech]

Applies for osu standard too
In my case appears to be related to the accelerated replay

2022-08-30.19-04-25.mp4

[Sobsz]

Another example in 2022.923.1-lazer: Sobsz playing EMILIA (CV_ Takahashi Rie) - Stay Alive (Seto Kousuke) [Insane].zip

In the actual play, I failed just before the end, but the replay fails 12 seconds before the end instead:

Additionally, skipping to the very end leads to a pass:

performance-audio.log
performance-update.log
runtime.log
updater.log
database.log
network.log
performance.log

[peppy]

Probably warrants investigating as these may be incorrectly being sent to the server too.

[smoogipoo]

I'm investigating this one.

Part of the issue seems to be the AllowedImportantTimeSpan check which exhibits different behaviours with/without fast forwarding:

osu/osu.Game/Rulesets/Replays/FramedReplayInputHandler.cs

Lines 113 to 114 in f318452

	return IsImportant(StartFrame) && // a button is in a pressed state
	Math.Abs(CurrentTime - EndFrame.Time) <= AllowedImportantTimeSpan; // the next frame is within an allowable time span

Another part of the issue seems to be slider start circles tracking slightly in non-ffwd when compared to ffwd. Likely the slider having one more update frame in the non-ffwd case.

The third thing to consider is that we're truncating replay frames to ints, so it's possible that replays can differ from gameplay.

[ekrctb]

Another part of the issue seems to be slider start circles tracking slightly in non-ffwd when compared to ffwd. Likely the slider having one more update frame in the non-ffwd case.

is tracked in #13352

[Detze]

The OP is just #11311 (comment) (map is OD 7, gameplay's Perfect hit window is 17.2 ms, replay's is 17.5 ms due to rounding, rest are all 0.5 ms wider in replay than gameplay due to rounding, so replay accuracy is better across the board - stable would be the same as replay, but fewer Perfects due to differing Perfect hit window formula).

[bdach]

closing as duplicate