The specification is a bit ambiguous about "notation data" signature
subpackets marked as "critical".
Implementors may assume that their implementations can safely ignore
such subpackets since they do "understand" "notation data". This
may lead to problems with human-readable notation data whose display
to the user may be critical for whatever reason (including legal
ones, think liability limitations), and it may lead to even worse
problems with notation data which is not human-readable.
How do the current implementations handle this situation (most
notably the NAI ones)? Should the specification be changed? Should
extensions to RFC 2440 which have to put critical information into
signatures define sub-packets of their own, or should they use
notation data to avoid name-space cluttering?