I have modified the Shavian wiki so that the metadata is held on article pages instead of talk pages.

Ooh, I think I like that.

Especially the Androcles information -- when I was entering lots of data, knowing what was canonical when I looked for similar words was very useful. Seeing an active talk page link was a good clue that the word was probably in Androcles (since most talk pages existed only to document this), but wasn't a sure-fire thing. Having that information on the article page is good, I think.

I am wondering whether it would be generally more useful if the data was held in IPA format and the Shavian text was produced using a transformation on that data, just as Unifon and so on are now.

What advantage do you see for this?

If there is a 1:1 mapping between Shavian and IPA (as, presumably, you would need), it doesn't really matter which is the underlying representation, does it?

And if it's not 1:1, then there will be IPA entries which cannot be converted to Shavian. (Or not correctly.) I'm not sure whether this is desirable for a Shavian wiki.

If it's a generic "spelling reform" wiki, then this may be less important; there will always be edge cases where automatic conversion doesn't quite work, if only because different reforms typically presuppose a different phoneme system. (Compare, say, Deseret with Unifon with Shavian.)