Over lunch, I was musing about the problem of change in conlangs.  When
you're making a language, if you're going to make any progress at all, you
have to draw lines in the sand and say "This is how it's going to be". But
sometimes you have to change those lines in light of subsequent rules, and
sometimes you change the rules to fit the lines.

== Begin explanation of the conlang problem which prompted this thinking,
which you can skip ==

There are two names in Nimyad, which have existed from the beginning of the
language in real time (in the 1980s), and which are anomalous under the
current phonology. They are Calsi, the name of the island on which Nimyad is
spoken, and my own name, Marnanel. They are anomalous for the same reason.

The etymology of both names is straightforward. Calsi means "(wisp of)
cloud", after its appearance from the sea (it's a chalk island). "Calsi" is
the zero-grade form of "calas", "thundercloud", plus the suffix "-i", "a
small self-contained piece of something". (Compare "tasi", lightning, from
"taras", light.)

Marnanel means "dreamer". The agentive suffix is "-el". "Maran" means
"sleep" (the same word means "fire" for reasons which make sense but would
be too much of a digression here), and we add "-an" to the zero-grade form
of a word to mean "the contents of something". So we have "marnan", dream,
which is both a noun and a verb.

Okay, so where's the problem? The current rules say that the zero-grade form
of "maran" is "man", not "marn", and the zero-grade form of "calas" is
"cas", not "cals". However, changing these two names is not up for debate,
so the rules are clearly wrong.

Fortunately, there is a pattern here: both are liquids following an "a". So
we could say that the zero-grade form of a word consisting of
C+"a"+liquid+V+C is C+"a"+liquid+C rather than C+"a"+C. We could further say
that the "ar" in "marnan" is an R-coloured vowel (which is after all how I
say the word when I'm speaking English anyway), rather than the vowel
followed by a flap.  This works, but would have knock-on effects such as
"tasi" becoming "tarsi". This may be worth the regularisation. We could also
just say that these two words are anomalies and be done with it.

Another difficulty is that Amlin, the conscript, is by design very bad at
representing consonant clusters; "marnan" can be represented, but it needs
carriers. It's a bit soon after it settled for modifying the alphabet. I
think it *may* be worthwhile to say that "marnan" is spelt in Amlin as if it
was M-A-N-A-N and that the colouring of the vowel is merely implicit, but
gets used when transliterating to English; alternatively, I can quite simply
superimpose some letter pairs, so M-A-R/N-A-N; alternatively, we can just
use carriers for this fairly rare case.

== End explanation ==

Anyway, I thought other people must run into this kind of problem fairly
often.  How do you deal with it, generally, if a word you're attached to is
anomalous under a newer rule?  Or do you not get attached to particular