pull down to refresh

There's bias correction techniques. The simple version of which is:

You can generate unbiased entropy from coinflips by flipping the same coin twice, and applying the following mapping:
  • Heads, Heads: Discard.
  • Tails, Tails: Discard.
  • Heads, Tails: Add a 0 to your entropy.
  • Tails, Heads: Add a 1 to your entropy.
You may also use dice, in which case you will roll a single die in sets of 2 rolls. Each roll, check if the number is even or odd.
  • If roll 1 is even and roll 2 is odd, record a 1 bit.
  • If roll 1 is odd and roll 2 is even, record a 0 bit.
  • If the rolls are both even or odd, discard the result
Do this until you have 256 ones and zeros. This is your entropy. [1]

The more technical and in depth version of which is:
https://wuille.net/posts/binomial-randomness-extractors/

  1. https://btcmaxis.com/article.html?id=7554e7cb-d8aa-45d5-95c8-adea8d87ea23

Yes absolutely, Von Neumann debiasing, which entirely nullifies first order bias. It kinda takes forever, though I believe you can go faster with a die doing: 1 if every other throw is higher, 0 if every other throw is lower, discard if they are equal. (I think Codex32 uses that from the top of my head)

It's a very good solution if you have the patience, probably closer to optimal than using tiles, but I was looking for the best solution that would be fast & simple. Btw, I'm curious, how do you compute your checksum if you go for 24 words?