Re: [openpgp] Mining protection in fingerprint schemes

[Daniel Kahn Gillmor <[e-mail]>]

On Mon 2016-04-11 08:04:20 -0400, Phillip Hallam-Baker wrote:
> Fingerprints are used for two purposes.
>
> 1) To define a trust anchor.
> 2) To verify that a purported trust anchor is correct
>
> Only the first of these requires the ability to enter the fingerprint
> at a keyboard. .

For my own practice, i've found that i do better verification by typing
in a fingerprint than by reading and checking in my own head.

if i'm asked to compare two fingerprints that are displayed to me on
different media (say, a computer screen and a business card), i observe
that i simply lose patience and focus before the comparison is
completed and tend to opt toward "yeah that's good enough".

But with the other approach, it's pretty easy to power through the "data
entry" task of transcribing the (relatively short) fingerprint from the
business card to the computer, and just let the machine do the
fingerprint comparison.

I have no idea how prevalent these particular cognitive biases are with
other humans, but i'd be reluctant to say that we should expect people
to always do verification of fingerprints without data entry.

> The most efficient way that is compliant with existing
> keyboards etc. is Base32. While not all keyboards have latin
> characters, there is an effective limit on the number of keys, screen
> real estate etc.

Here is a link to base32 encoding for folks who aren't used to it:

  https://tools.ietf.org/html/rfc4648#section-6

a base32-encoded bitstream should be marginally smaller than (it should
be 80% as long as) the same bitstream encoded as hex.  A 40-bit field
would consume 8 octets in base32, but 10 octets in hex.

i would really like us to make a clear recommendation for a common
textual interchange format for these identifiers; i'm not convinced
that it's actually out-of-scope for 4880bis.

     --dkg