Proof-or-Stop: Don't Trust the Agent, Trust the Evidence -- Loop Engineering for Verifiable Evidence-Gated Lifecycle Control

arXiv cs.AI · 6 hours ago

Researchers present Proof-or-Stop Lifecycle Control, a method that requires autonomous coding agents to provide verifiable evidence before transitioning between software development states like tested or ready-to-merge. The gated approach reduced false-pass rates from 31 of 1,800 injected test cases to 2 of 1,800 in controlled ablations. The system treats agent outputs as claims requiring proof rather than assuming they are correct, enabling safer autonomous development workflows.