Found it! The setting was "Preserve spaces"
The txt files were using hex value '20' as a space, which if "preserve spaces" was off got converted into 'C2 A0'
Both the K3 and Calibre viewer show this as a blank, but don't seem to treat it as a valid character with which to separate words, so they treat the entire paragraph as a single word.
This also explains why it stuck through conversions, as the '20' space was still present to be carried over to the next format.
Thank you for your help in this. I don't know if you want to treat this as a "bug", as the setting was preserving the spaces as they were.