Reproduce “Proof-or-Stop: Don't Trust the Agent, Trust the Evidence -- Loop Engineering for Verifiable Evidence-Gated Lifecycle Control” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jek Huang, Jeffery Hsia, Jiayi Sun, Freddie Shi, Wei Huang, Ian H. White
Jek Huang et al. at Prodenovo created Proof-or-Stop, a control framework that admits autonomous software lifecycle transitions only when verifiable, code-state-bound evidence supports agent-emitted claims. This mechanism works to reduce the amplification of hidden defects and propagation of unsupported software artifacts.
Key insights
- Autonomous software development should operate on 'agent-as-claim' semantics, where agent outputs are propositions requiring verification by admissible evidence, rather than direct assertions of lifecycle state.
- Mechanically binding evidence to specific, cryptographically sound hashes of the exact tracked source state (material, head, story files) ensures freshness and integrity, defeating stale or tampered claims.
- Host-neutrality is characterized by the ability to verify local evidence against an identical code identity, enabling verifiable cross-machine resume and mitigating implicit trust in remote hosts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.