Precedent as permission

What it looks like. "We've done this before" is a sufficient argument. Overrides exist for emergencies and are used weekly. Anything not explicitly forbidden is available. What the system actually does has drifted from what policy says, and practice wins every time they disagree.

Why it fails. Each step is small and locally defensible, the ratchet only turns one way, and after enough steps the operative boundary is somewhere no one would have approved in a single decision.

Instead. Anchor normative boundaries to canonical text and treat precedent citation as a reason for more scrutiny, not less. Track invocation frequency per override path against a hard threshold. Replay policy against actual practice periodically and force an explicit choice: change the policy or change the behavior. State policy-space closure — open-world or closed-world — per domain rather than leaving it implicit.