Well, in the first instance, obviously it was dumb.
However… I would shoot a interview without looking through the viewfinder if I checked it on live view. When I recorded in stereo mode I did visually check the input levels, i.e. that they were rising and falling with the conversation.
When I recreated the scenario of recording in stereo mode and selecting the XLR inputs (i.e. with no mic attached), there’s no input levels shown (obviously). That’s why I can’t understand why I ended up recording silence. Either I’m blind, my memory is faulty, or there’s a third explanation.
Anyhow, lesson learnt.