Single host task groups aren't guaranteed to run on the same host. Files created locally and required for later tasks will result in flaky failures.
This is a somewhat rare issue, but I was able to find about 12 occurrences in the last 60 days and a linked BF ticket from Jan 2024.

The fix for this is to just trigger a task restart when the files are missing. Evergreen documentation states that restarts on tasks in single host task groups will result in a restart on the entire task group. The retry_on_failure will only restart the task once according to documentation, so it shouldn't spawn an infinite loop of task restarts. There is also a maximum amount of automatic restarts allowed on the evergreen project so this could impact the quota, but I'm expecting the usage of this to be so minor that it won't make a meaningful impact to that quota (200/24hrs).

Found during https://jira.mongodb.org/browse/DEVPROD-42457