Met with JJ and JT. Results: I've overwritten the current data/text versions of the test plays with my own generated versions, which now have glyph encoding; in the case of H5, I actually ran the glyph processing on the existing versions since JM confirms he's been editing those directly (although not since July). JT has brought back some of the collation/annotation offset-checking code from GitHub so we can integrate that. We've decided that we should not hard-convert editor-specified offsets using string matches into anchors until rendering time, because it's easier for editors to work with them if they're text-based, but we'll provide easy checking for editors.