Version -06 includes fixes for the typos, and is also more explicit about the
fact that certain fields are included only if the corresponding flags are set in
the initial word.

The object syntax still allows 'meaningless' MRIs to be constructed, for
example, setting both S (SPI) and A/B (port) flags, or setting S when the
protocol number is not one of the IPsec protocols. However, rather than trying
to pin down exactly what combinations are acceptable in the bit-level syntax, it
seems more reasonable to rely on 
a) common sense, and
b) helpful error messages which can be generated by an node receiving a
combination which it regards as unreasonable.

We should also consider whether the port fields should only be allowed if the IP
protocol number is something sensible (UDP, TCP and so on).

The text should also say that if the A/B, I flags are not set, the corresponding
words are absent in the body of the information element. (This is partly implied
by the text for A/B already.)

... and the text about one or other of A/B should really refer to A/B and not
S/D (which is old terminology...)

It would be helpful to clarify that if one but not the other A/B flags is set,
the bits in the ports field for the 'unset' port are treated as padding (i.e.
present but not significant).

more generally, it would be helpful to verify (and then state) that the parsing
interdependencies between different fields only affect the flag fields in the
initial word. once the set of flag fields has been checked for consistency, the
rest of the words in the object can be parsed independently.

Now that "S" is no longer used as the Source Port flag, it might be logical to
use "S" instead of "I" for the SPI.

It is hard to think of a circumstance where it would be meaningful to specify
both of these, so the bit format should probably be fixed.

More generally, the flow-ID format needs to be checked to ensure that it
actually correctly reflects the set of possible meaningful combinations of
header fields that can be specified.

The current ABNF for the flow ID (MRI for path-coupled routing) disallows the
specification of both SPI and L4 ports. But the bit formats in C.4.1 allow both
at the same time, which is a contradiction.