Energy conversion balance is slightly off in a supposedly lossless converter

This seems a really interesting matter (no sarcasm)
I was doing some power losses analysis and found out that my input, losses and output powers did not check out. So I dug in the rabbit hole and after some experiments it seemed to me that the LC tank, which does not have to my knowledge any lossy element in its components model, was exactly where the missing (average dissipated) power was hiding.
So I built this extremely basic experiment to rule out any possible other contribute.
Sure enough, averaging the difference of input and output power over a finite and integer number of periods does result in a non-zero quantity.

I was thinking about a couple of explanations, that I don’t know how to verify, but I would like to hear from the community first.

Limit MAXSTEP, this delta seems to be fine. You are working on square wave, more data points may be better.