Why Passing All Internal Checks Still Can't Guarantee a Correct AI Agent Handoff

A software developer demonstrated that an AI agent handoff artifact can clear every internal verification check and still contain factually wrong data. In a test fixture, seven checks — covering structure, identity, provenance, constraints, conflicts, and encoding agreement — all passed, yet an external receipt check failed because the artifact referenced the wrong authentication provider and an incorrect database row count. The core issue is that internal checks like schema validation, provenance tracing, and cryptographic commitments only confirm internal consistency, not correspondence with real-world state. The developer argues that 'verified' should be disaggregated into separate layer-by-layer results rather than a single pass/fail verdict. To illustrate the concept, they released an open-source tool called Babel Context Integrity that replicates the wrong-branch failure scenario offline.