Prior to calling the scheduler, save each individual JobSpec object to the scratch directory. This provides a useful debugging checkpoint, and opens up a number of potential new features... such as pausing and resuming regressions, retrying failed jobs due to transient issues (license availability).