do you mean having validation or cleanup (setting 4 bits to 0) that makes sure that all bytes in the struct are actually nibbles?

Yes, but it's impossible if we give out mutable access to the bytes directly (&mut index)

It's currently discouraged to have invalid state but not unsafe nor impossible

I think we either:

1. don't clean nibble input but indicate that methods are for changing nibble content and that would be a caller responsibility (current state)
2. clean nibble input when inserting (e.g. v & 0x0f) but that should be done for set_at, push and extend_from_slice for consistency.

what do you think?

1 is ok for now, I think I want to revisit this in the future