Yeah, it was an authentication issue. Turns out the compute clusters were set up with instance profiles, but never the job profiles, so when the wheel process was set up it failed for jobs.

TL;DR: you need to apply instance profiles for access to shared resources on the compute cluster.