If it's any consolation to anyone, I haven't been able to recreate this in Windows 7.
Background: I have a solution in a Windows environment with some very similar scripting. I pulled down a backup from the FMS12 server and opened the file locally in FMPA 15 (W7 WS). I ran multiple import runs, with multiple Quits/ Exits in between, to try and trigger caching behaviour.
I also adjusted scripting; in my normal one, the entire procedure happens within a single script whereas in yours, you appear to have an Exit after the deletion(s) from the Import Table. I also added this structure eventually, as nothing else seemed to trigger adverse caching behaviour. Even that worked perfectly.
Note: besides the completely different OS, the local FMPA is 15.02.
Thanks for taking the time to test, Mardi.
Have you considered the possibility that you're cross-talking with another version of the file (hosted perhaps) that still has 7023 records in it? What do your external file references show for the path?
I think that was it, Chris. The import script step had a legacy File Path. I added another file path and although they look the same (see image), they apparently were not (I removed the top file path before re-testing). Once I re-specified the File Path for all the imports, it brought in 101 records, as expected.