Why does the bullet rating distribution look like this

lichess.org/stat/rating/distribution/bullet
it has spikes, I was trying to think of a reason why it's like that but I couldn't think of anything, what do you think?

It turns out the same is for ultrabullet, but longer time controls do not follow this, lichess.org/stat/rating/distribution/ultraBullet please explain why ty

"Oh man, my rating is almost 1700, I'll play a few more games to see if I can reach 1700."

"I just reached 1700, great achievement, now to go do something else."

It is all psychology, while people can't decide how their rating is going to fluctuate, they may decide when to stop playing based on rating, when enough people do so, and there is a general agreement about what ratings are keep-worthy you can see it in the statistics.

For longer time controls you typically play fewer games per session and have less flexibility in when you stop playing, so the effect is lessened.

@NohatCoder's prediction seems likely for the bullet distribution, especially considering that most of the spikes land on multiples of 100.

For the ultrabullet distribution, it may be due to the relatively low number of players accounted for. With only about 4000 players, the graph looks very rough. It might be more likely to even out with more players, but that's just speculation. It might end up the same way as the bullet distribution.

All of the variants seem to have similar spikes to the ultrabullet distribution, although seemingly more pronounced. This makes me think it might be due to low player counts. This doesn't apply to bullet, which has about 76000 players weekly.

This needs to be answered by lichess. What’s going on here?

@NohatCoder is absolutely right, therefore the case is already solved @GMScuzzBall

@Mollus no way that’s true for so many. It must be something else.

@GMScuzzBall

You've committed a Personal Incredulity Fallacy: Just because it seems extremely unlikely to you that people would stop playing once reaching a multiple of 100 doesn't mean that it is impossible.

That being said, everything in this thread is still speculative, although @NohatCoder's thoughts seem likely. Nothing here is proven.