Turns out a #kubernetes `emptyDir` volume mount isn't cleared for pod restarts due to crashes.
I'll remember to always clear my scratch spaces in the service.

Seems like something the infrastructure could take care of though.