You got it right! both happens. The integrity gates at 2.5 and 4.5 run a 7-mode checklist and hard-block on serious stuff like hallucinations or unsupported claims. Milder flags just warn and let you decide at the human checkpoints.

→ View original post on X — @alphasignalai