This is a bit backwards as I've already submitted the pull request #30 for this, but just for tracking the issue:
We were reading raw 16-bit integer values from the DAQ card, which is not advised with current NI DAQ cards because this value has not been scaled to reflect calibration. For the most part this only introduced a slope error in the voltage - to - digital value conversion, so we didn't notice.