I'm finally releasing some new PPU VBL/NMI timing information and test ROMs. These tests brutally test to single PPU clock accuracy, unlike previous PPU timing tests, so there's no room for error here. Please note that these test only one of the random timing arrangements at power-up (there are others with more special cases, but I haven't spent time determining them).
Which test does that occur on? My guess is that somehow one of the NMI handlers is firing way before I expect it to, which is causing it to use an uninitialized error code. I'll have to add a fix for that to my test framework.
Thanks for the info. I just checked and the first two tests in ROM 1 don't properly set the result code, so if they fail they'll print an erroneous code as you encountered. I've updated the original archive with this fix.