When a decision should stop carrying weight.
Not every AI-supported decision deserves the same level of reliance forever. These are the plain conditions under which a decision should stop being relied on silently — not an execution rulebook.
A decision should not continue silently once its basis has changed.
Stronger reliance on a decision needs stronger support behind it — not just continued use.
High confidence in an output does not substitute for missing evidence.
Analysis, policy and legal conclusions must not be silently blended into one undifferentiated finding.
Every decision needs a visible reliance boundary — where it stops being usable to support further action.
This is not an automated kill-switch specification, and not a public description of how SnapOS's private engine decides to halt anything. It is the set of conditions worth recognizing before continuing to rely on a decision — the judgment itself stays with the people accountable for it.