Investigation of the ROC31 problems found that the VQWK gate inhibit signal which is asserted at the beginning of the run was inverted (should be on output #7 of TIR with yellow wire to the right).
Also, a number of samples setting of 925 seems to be too large for an IT setting of 2920. I changed the number of samples per block to 905.
Calculation for nsamples:
integrate_time = IT * 2.5 us = 7300 us integrate_time - VQWK delay = 7300 us - 42.5 us = 7257.5 us (integrate_time - VQWK delay)/nblocks = (7257.5 us)/4 = 1814.3 us nsamp = (integrate_time - VQWK delay)/nblocks/(2 us) = 907 samples
I dropped it down to 905 for now.
I am not sure why entry 324477 indicates that we had been able to run with a setting of 925 earlier, but not now. I would guess that the difference is in the T_settle period, which is now 100 us, and may have been 500 us when we were running with 925. A setting of 925 leads to a VQWK response about 160 us longer than 905, so we might have been pushing into the T_settle period earlier and now we're running into the next T_stable.
Anyway, the PREX run type seems to work now, with these settings for the VQWK modules: