The Implementation Lottery: Auditing Idea Reliability in Automated Research.

arXiv:2607.26587v2 Announce Type: replace-cross Abstract: Automated research agents use program scores to judge ideas. We call variation in this evidence across implementations the implementation lottery. We introduce an Idea Reliability Audit that freezes mec