A false-positive completion is a feature an agent reports as complete that later fails review or testing. It occurs when completion is accepted on the agent’s claim rather than demonstrable evidence. Imprecise success criteria can also let an incorrect solution pass. section 3.4.2 section 2.1.2
Use false positives as a signal about whether intent is verifiable:
- Observe them qualitatively at first; a high share suggests the completion gate is relying on claims instead of proof. section 3.4.2
- Track them alongside specification bypasses and alignment regressions to keep the specification and implementation aligned. section 3.4.2
Listing 2.1 A unit test with weak assertions in pseudocode
test "filters active users"
# Arrange
users = [
{ name: "John", status: "active" },
{ name: "Jane", status: "inactive" },
{ name: "Chen", status: "active" }
]
# Act
result = filterByStatus(users, "active")
# Assert
assert result is not emptyA test that checks only that an active-user filter returns something can pass even when its results are wrong. section 2.1.2

























