Observed arrival · 2026-11-01
ProofGate: checking what an AI agent’s “pass” actually means
AI Evaluation Lab presents ProofGate, an open-source browser demo for independently checking recorded coding-agent patches against a task contract.
- For
- Engineers evaluating coding agents
- Worth noticing
- The demo runs six fixed controls, including a known-bad baseline and four deliberate mutations.
Field notes
The page lays out a verification sequence: define a bounded task contract, challenge the verifier with reference and deliberately faulty candidates, then retain code, failed checks, source references, and execution reports. Its recorded AP-005 example says public tests pass while independent checks flag four issues, including row preservation and input integrity. The site also cautions that results apply to tested tasks and conditions, not general safety certification.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue