Classic scores set on some beatmap appear to have its total score heavily inflated.

[DarkbitNetwork]

Type

Game behaviour

Bug description

For some reason, this beatmap seems to have all classic mod plays set on it to be heavily inflated. The top 4 have total scores that are higher than what should have been possible in lazer. Every other classic mod score also are affected though a bit less extreme.
This bug can also be seen on the website.

However, it seems like non-classic lazer plays are unaffected. I don't know if scores with classic mod set on lazer are also affected by this bug.

A downloaded replay of the inflated score seems to fix the total score though.

Screenshots or videos

Top 4 have a total score that's higher than the expected limit:
1

The downloaded replay of a top 1 score:

Version

2024.906.2-lazer

Logs

compressed-logs.zip

[bdach]

A downloaded replay of the inflated score seems to fix the total score though.

This is rather worrying because in theory the score conversion process client-side and server-side should be the exact same process executing the exact same code. #29813 is another case of this.

@peppy can you please check on the specific version of the osu.Game package used by the osu-queue-score-statistics stable score import job? I worry that it might not contain some of the scoring conversion fixes that I've done at the start of this year.

[peppy]

@bdach pppy/osu-queue-score-statistics:2024.706.0, last restarted 49 days ago. Let me know if you need more information.

https://hub.docker.com/layers/pppy/osu-queue-score-statistics/2024.706.0/images/sha256-a621c7af15fa8b4b454599ce34fed2d3915d0225b70f08611aaee78e102de253?context=explore
https://github.com/ppy/osu-queue-score-statistics/blob/dafc9b4b39c7e205c768d0fb7b7324f70aa1e54f/osu.Server.Queues.ScoreStatisticsProcessor/osu.Server.Queues.ScoreStatisticsProcessor.csproj (in theory)

This is specifically the high score importer, FWIW. The non-high score importer is older, sitting on 2024.312.0 (probably should be bumped).

[bdach]

Okay... then maybe it's the other end? Where's osu-difficulty-calculator at, specifically the job for computing scoring attributes? (I believe that might be a separate process from diffcalc, but not sure)

[peppy]

pppy/osu-queue-beatmap-processor:latest (https://hub.docker.com/r/pppy/osu-queue-beatmap-processor/tags)

..so https://github.com/ppy/osu-difficulty-calculator aka https://github.com/ppy/osu-difficulty-calculator/releases/tag/2022.928.0

[bdach]

That cannot be true, scoring attributes did not exist in 2022.

I am asking about scoring attributes specifically, e.g. whatever it is that is populating osu_beatmap_scoring_attribs for new ranked maps. The thing from ppy/osu-difficulty-calculator#224.

[peppy]

This is what's running on kubernetes, if it's helpful:

[bdach]

Hmm. latest is https://hub.docker.com/layers/pppy/osu-queue-beatmap-processor/latest/images/sha256-b146bf6bfcd7ec6204aa9d7c8ff98f0175a5649c03e86738e5ef80987e87c005?context=explore which is dated Dec 18, so probably somewhere at ppy/osu-difficulty-calculator@b8fa416. Which is pretty old and possibly causing this. The theory would be that while we did deploys of scoring attribute changes after my fixes previously, the processor got forgotten and as such rolling scoring attribute updates are using a stale revision. I probably should have specifically reminded about that...

One last thing: what does

SELECT * FROM `osu_beatmap_scoring_attribs` WHERE `beatmap_id` IN (4713870, 1593298) AND `mode` = 0;

spit out? I'll use the output of that to crosscheck scoring attributes and see if that theory checks out.

The good slash bad news is that if it is that, we have all the means to correct this, it'll just take time...

1. Update the processor so that it's using newest game packages
2. Recalculate all beatmaps' scoring attributes again to cover the gap that was using old code
3. Reverify all scores

Maybe it'd take less time if we could perform steps (2) and (3) selectively but I'm not currently sure we have the means to do that. (3) maybe could be done selectively, if we can conclusively somehow get a known-good starting score ID.