Run a post-generation self-audit tailored to the dominant failure mode (loopholes, fake precision, imported constraints, too slow/too many questions). Use when you need a reliable “second pass” that actually catches the common ways prompting guidelines backfire.
Check:
Check:
Check:
Check:
After producing the output, run a targeted self-audit:
Scan type: <loophole|fake-precision|imported-constraint|too-slow>.
List what you found and revise the output to address the top issues.