It is difficult and time consuming to achieve a high confidence that a die is not biased.
By rolling a die and performing a statistical analysis you may be able to quickly tell that a die is very likely biased, for example if your first 10 rolls are a 6 there is only a 1 in 60 million chance that the die is not biased.
Instead suppose you roll (3,3,8,3,1,2,3,3,2,7). Half the 10 rolls being a 3 tells you little, it is likely that you would need to perform many more rolls before having confidence tha the die is biased.
You can't quickly tell that it's not slightly biased.
When generating a entropy for bitcoin purposes (BIP39 seed) it is safe and generally easier to increase the number of events than it is to interrogate the entropy generated per event.
So, we will make many dice throws.
What if you could roll 100 dice in one throw?
You can download the STL file here. I'd love to hear if you print one, tag me @orangesurfbtc.