Some of the CSL styles for Zotero use the UTF-8 non-breaking space character. For instance, the American Physics Society's in-text citation form is a number inside of square brackets, preceded by a non-breaking space.
All of the strings are sent from Zotero to TeXmacs tm-zotero.scm as UTF-8 strings in JSON format. Before it can successfully parse the LaTeX that the citeproc-js outputFormat I've created for this, 'bbl', the LaTeX string must be converted from UTF-8 to Cork. That is because the parse-latex function does not have a way to tell it to use UTF-8 encoding; there are heuristics in that routine for determining if the input is UTF-8, but it's not easy to fool it and so I convert to Cork encoding.
The problem is that what I'm getting back from it is the literal string "<varspace>", which is not being displayed as a space. I think there's a regression in TeXmacs causing this.