Hi Everyone,

Thank you very much in this thanks giving day for the detailed and well thought out responses. :)

Steven Hatzakis via bitcoin-dev <bitcoin-dev at lists.linuxfoundation.org>:
> *Option 2*: Perhaps a revision is needed to how the BIP39 seed is
> generated in the first place, such as by hashing the entropy instead of the
> words. Any thoughts on how viable that could be where the initial entropy
> is fed into the PBKDF2 function and not the words?
If we go this direction, I'd suggest that we pull Shamir's Secret Sharing into the game. Trezor's 
SLIP-0039 proposal is great and has many security aspects already covered. However, it does 
not allow any language other than English and Trezor team clearly stated that no other language 
will be supported.

While I really want to keep the language independent design. So in the revision, I'd like to see 
a language id (allocated to each one having a defined wordlist) in the SSS share, as well as
share id, threshold, index, share value, checksum etc. 

Regarding checksum scheme, SLIP-0039 proposals a 3-word Reed-Solomon design. It has a very
good error checking capability but not very good at providing hints to error recovery. Trezor team 
opposes to the idea of providing hints to users regarding how to fix an error. This could lead to
difficulties for some vendors, and in small probability, confusions to users (when there is a 2-word
error)

I do agree with Trezor team that it should be users' responsibility to recover from a detected error.
However, there is a better way than solely rely on checksum. That is, as in our revision, we can 
support mnemonic in multiple languages simultaneously, why don't we use two languages, or one 
language + numbers to check each other? In Steven's example (language id, share id, etc. skipped) 
we could record a SSS share (assuming it is one of the shares just for the sake of example) like:

> *In English*: minimum fee sure ticket faculty banana gate purse caught
> valley globe shift
> *In Spanish*: mercado faja soledad tarea evadir aries gafas peine búho
> tumor gerente reja
Or
> *In English*: minimum fee sure ticket faculty banana gate purse caught
> valley globe shift
> Word Indexes: 1128, 676, 1744, 1805, 653, 145, 770, 1396, 291, 1927, 794, 1582

Then software will have to check checksum as well as to check if words match each other. For 
example, "minimum"'s index value in English wordlist should equal to "mercado"'s in Spanish,
or should equal to 1128. 

If any error is detected, combining the checksum value and dual-encoding information, it is much
easier to figure out which word was handprinted incorrectly. 

BTW, it is very error prone to handprint. Some study suggests about 0.9% per word rate. See
http://panko.shidler.hawaii.edu/HumanErr/Basic.htm

Hotopf [1980]

W sample (written exam). Per word

0.9%


It is important to have an error recovery mechanism easy to understand and implement. 

Thanks,
Weiji