It was observed that there are tiny discrepancies, from time to time, between the transfers stored in the csv file (currently in units and not atoms) and the actual transfers in the calldata. Amount differences are insignificant, but this still blocks a more automated way of verifying the correctness of the calldata.
Note: a first attempt to simply store the amounts in atoms in the csv and not scale down to units still doesn't seem to work