All we really do currently is check if the block entries are distributed in some way. That is reasonably clear by construction and I dont think we really need to invest pytest runtime into verifying that...

Maybe we should test the proper variance once per block-backend though, i.e. test only the methods that generate random blocks?

I think we should rather test / verify if the tensors themselves have the expected distribution (this assumption is baked into the implementation, and I am only 95% sure it is actually valid).

We could e.g. estimate some expectation values like and <T @ T.hc>, <T @ T.hc @ T> and so on by sampling and compare to the values that we expect. (btw: is it even clear how to integrate over the measure for anyons...?).