I did quick fiddle with the data in Excel. It looks like an offset of -45 counts and then a linear scaling error of -2.4% over the range of the input.
Fluke(ma)
Actual Count
Expected Count
Count Diff.
Data without offset
Percentage Error
4.000
-45
0
-45
0
6.000
954
1024
-70
-25
-2.44%
8.000
1954
2048
-94
-49
-2.39%
12.000
3953
4096
-143
-98
-2.39%
14.000
4957
5119
-162
-117
-2.29%
16.000
5951
6143
-192
-147
-2.39%
18.000
6951
7167
-216
-171
-2.39%
20.000
7949
8191
-242
-197
-2.41%
I would not expect this as normal behaviour. The spec sheet gives accuracy of 0.4%. However, apart from the offset and scaling errors, the data looks quite clean. Usually a bad input won't even perform this well.
Were you able to test with the field wiring completely disconnected from the analogue module (that includes the commons as well as the signal lines)?
I notice you are a few versions behind on UniLogic, the current version is 1.28.34. I doubt the issue is version related, but I would suggest upgrading at some point. I'd be pushing into a more thorough diagnosis of the hardware and field wiring first.
Hope this helps,