Advisory, then authority
A new guard ships with no power to act: the wire that would let it touch a reply doesn't exist yet. It earns that wire with weeks of measured false positives, not with promises.
New guards deploy advisory: they observe and log, and the code path that would let them replace a reply simply isn't wired. Only after weeks of accumulated false-positive data clears the threshold does the wire get built. Overreach during the trial isn't forbidden, it's structurally impossible.