[Sofi_blackbox]

Follow-up: This test shows that LLMs sometimes continue producing when any output is illegitimate under their own accepted rules—exactly the scenario my SOFI framework highlights.