For example, try adding -Q (reference source coupled after frontend) to the vdo12 MEM command
pcm -d ../cal/database.txt -s -c calib.TT *.zz -Q
If you filter out the FluxCal observations,
grep -v -F Flux ../cal/database.txt > ../cal/database_no_fluxcal.txt
and re-run the pcm command with this database file, the command finishes successfully.
I suspect that the problem lies in how the FluxCalManager adds a new SingleAxis backend for each FluxCal observation, but only the off-pulse baseline is propagated through it.