Figure 12.2: The four consequence tiers, keyed to what an action costs when it is wrong and never to the model’s confidence
Figure 12.2 The four consequence tiers, keyed to what an action costs when it is wrong and
never to the model's confidence. Read-only actions run freely; reversible ones run
provided they are logged and can be undone; externally visible ones earn a review queue;
and irreversible ones (in accent, at the top) wait for a signature, every time.
Consequence rises up the stack to the one gate you cannot afford to skip.
