Scheduled jobs often work on files: a nightly import, a weekly report template, a price list. If the file disappears between scheduling and running, the job fails for a reason nobody can see.
Declared inputs are retained
A scheduled job can now declare the files it reads. Those files are retained for as long as the job exists, through every run and every component that uses them. Cleanup skips anything a live job still needs.
Released when the job ends
When the job is cancelled or completes its final run, its hold on the files is released, and normal retention applies again. Nothing is kept forever by accident.
Checked on real databases
We exercised the change against several real database histories before shipping it, including restore paths, so retention survives a restore as well as a restart.