possible bug in chess insights?

When displaying average centipawn loss by date, and filtering for the last 6 months, I have the following odd behaviour:

When further filtering for variant and enabling/disabling blitz and correspondence, the displayed average centipawn values are highest when enabling both Blitz and correspondence.

The average should settle between the averages of Blitz and Correspondence but not exceed both. Or am I math-blundering here?

Yes, I noticed the same thing.
My Bullet ACP way lower than my classic/blitz, which doesn't make sense to me.

While this is odd, it is not contradictory in your case...
But an average of two weighted averages never can be more than both values...

Yeah, I am on an old laptop, which has bad trackpad. I try to premove almost every single move and most of them of fail.
But Chess inside still shows that my bullet ACP is lower.
Bullet 35.8
Blitz 62.7
Classical 63.4
Correspondence 59.6
Crazyhouse 129.5

For #1, by adding or excluding games, you will change which ranges are used for the dates, which would make direct comparison difficult.

For #2, I'm not sure, maybe you have very few bullet games analyzed? It shows the sample size (number of moves) next to the ACPL.