Which also means you can diff the code output from two runs of the system and determine if any of the diffs amount to regressions in the code.

Developers and machines that can't explain themselves end up out on their ear or damaging the projects they're involved in.