Daniel O'Donnell wrote:

> tei:choice should accept all tei elements as children, right?

er, where did you get that strange notion from?

Go to
and you will see that the declaration reads

element choice {, ( model.choicePart | choice )* }

Members of the model.choicePart class are listed if you click on the 
model.choicePart link: they are
	 abbr corr expan orig reg seg sic unclear

2) Sometimes a single runic letter represents sounds spelled by two
> characters in roman script (especially diphthongs). Is tei:c the right
> element to use for what you might argue is some kind of ortho-phonemic
> translation. 

c is intended for single characters, but it is syntactic sugar for any 
kind of low level segmentation.

It isn't quite orthography to phonemics since we are going
> from one standard period spelling to another, but it isn't character to
> character either. I imagine syllabics to roman translation would have
> the same issue.

it might be clearer to make up your own tag for the syllabification, if 
that's what it is.