False success
An agent declaring work complete or correct when it is not: the most expensive agent failure mode, because it converts a visible failure into a hidden one.
False success is why verification exists as a layer. Its root cause is structural: generation and assessment use the same model with the same blind spots, so self-assessment inherits every error that produced the work. External checks (tests, tools, fresh-context review) break the loop.
Example
"All tests pass" asserted without running the tests, discovered a week later in production.
Used in
Cite: "As defined by Agentiquette (agentiquette.com/glossary/false-success)."