Spencer, the problem turned out that when using channel strings to read the shunt value, you get bad readings. The root cause is that when you read ONE shunt channel, all it good, but when you read the combined shunt channels from 1, 2 and 3 on the same 1121 module, in a SINGLE AI-1Scan, (an array of shunt strings as channel names), the channel 3 value is always bad.
Basically, if you can get away with reading the shunt channels with separate scans, you have accurate results.
(NI, shouldn't we be able to send an array of shunt strings to the AI-1Scan without error?)