Why a snapshot refresh is not an AI change
A practical method for keeping baseline facts, detected field differences, and editorial updates separate.
A baseline answers what is true at initialization. It does not prove that a user-relevant field moved between two admitted observations.
A change event requires the same canonical object and field, an explicit before value and after value, source evidence, and separate detection and effective times. A snapshot identifier changing is not sufficient.
AI3Radar therefore fails closed when a source only says snapshot_changed, drift_detected, signal_updated, or coverage_changed. Those records may remain useful upstream diagnostics, but they do not enter the public change feed.
Official provider releases and historical events remain useful content types. They are labeled as official updates or timeline evidence rather than rewritten as a before-and-after change.
This boundary reduces feed volume, but it preserves the meaning of change and keeps evidence inspectable.