Non-translatables inside glossary entries aren't colour-marked. I mean: when a a longish glossary entry encloses a non-translatables, the last one isn't coloured purple.
Is this what we want?
Oops, it's a little more complicated than this. I've a three-word expression that I've added as one non-translatable. Then the above is valid.
I'll remove the three-word entry from the list of non-translatables and add the three words separately.
I have now removed the three-word entry from the list of non-translatables and reloaded this list. Now I can see what's happening: the first and the third word of the expression aren't present in any glossary: they are coloured purple (which I prefer, since I find the info that a word is a non-translatables more important–others may have other needs, of course). The second word is in the list of non-translatables, but also in the glossary. It's coloured orange:
After I removed the three-word entry from the glossary, I notice that it's now (correctly) recognised as a fragment from the memory. However, the colour is green now, instead of purple.
When I enter a new segment, the three words are marked purple shortly, then the fragment match is detected and the colour green is overlaid.