Keep in mind that a truly fair die will still roll hot or cold over short periods. For 100 roll trials, about 50% of the trials are 10% or more off from the expected values, and off by 25% or more at least 25% of the time.

So, if you want to catch a weighting of 10%, you are going to need substantially more than 100 rolls. I ended up with around 800 rolls on each red die, and am only 90% confident in about half my calls (discard or keep)

I saw one study comparing 2 brands of D20s and they did 10,000 rolls each die. The more sides you have the more rolls it takes (i think the term is "degrees of freedom") part of the problem is that (and this is what a buoyancy test is good for) the imperfection might not be specifically favouring one exact face, you might get a data set that looks off but not landing on 1 face. I found 1 die that was heavily weighted to one entire half. When you analyze the roll data you have to consider adjacency of faces not just prominence of 1 particular face.

I think I might start a roll trial with a die I know is crap via buoyancy test and do a high roll-count test and compare to a die that's good or acceptable.

The universe hates you. That's why your dice hate you. That's why you should hate your dice. You should hate the universe too. In fact, if something hates you, you should hate it right back.

There's a lot of hate floating around. Hates floats.

Have I mentioned the hate yet?

Hate.

Reckless Hate Makes Baby Obi-Wan Cry.

The universe hates you. That's why your dice hate you. That's why you should hate your dice. You should hate the universe too. In fact, if something hates you, you should hate it right back.

There's a lot of hate floating around. Hates floats.

Have I mentioned the hate yet?

Hate.

I prefer "Omniloathe".