Using Base 21 for Human-Proof Unique Identifiers
        ================================================

   [[[ Update 2012-11-15: https://en.wikipedia.org/wiki/Base32
       describes a similar system; thanks to dkg for the ref. ]]]


Users must sometimes manually transcribe the unique identifiers we
generate.  Because certain digits and letters look alike (depending on
the font), users often mistranscribe, as the identifiers are just
random sequences with no inherent meaning to humans.

We can reduce the mistranscription rate by putting similar symbols
into equivalence classes.  The consuming software would treat all
entries in a given equivalence class as identical, so that even when
the user mistranscribes, the identifier would still work.

Each column in the diagram below is an equivalence class, giving a
"base 21" alphabet from which to generate safe identifiers:


            8             7                       3
          4 6         9   1       0     5         2
          a b c d e f g h i k m n o p r s t u w x z
          A B C D E F G H I K M N O P R S T U W X Z
                      Q   l                 v
                      q   L                 V
                          j                 y
                          J                 Y

For example, if the original identifier were "9sxA3y", the software
receiving it would accept "qsx43y" too (for that matter, it would even
accept the much further off "G5X4zu", though one would have to wonder
about the user in that case).

Base 21 is a high enough base that identifiers will remain short for
most uses -- with 10 places, we get 16,679,880,978,201 unique values.