Turns out I’m not seeing the same behavior as before, where some errors were in different places with respect to the “Processing…” messages.
Curiously, after the last “Processing…” message, when the stuff in modularize at the end is checked, the order of the errors can vary from run to run, so I guess there must be some multithreading going on or something, to make it non-deterministic. But it doesn’t seem to be an issue. Sorry to trouble you.