Glossary · Term

Chain-of-Evidence

← all terms

Definition

Plain language

A standard saying every claim in an AI-generated paper must be traceable back to a real source.

As stated in the literature

A verifiability contract for autonomous research agents requiring every citation, number, method description, and conclusion to point to a recorded grounding artifact (paper, log line, or code), framed by analogy to ACID guarantees in databases.

Also called: CoE, CoE Integrity Audit

Why it matters: It's what separates a trustworthy autonomous research agent from one that confidently fabricates plausible-sounding citations and numbers.

For example, a claim that 'method X improves accuracy by 3 points' must link back to the exact log line and code commit where that number was produced.

Heard on the show

“… The paper is "ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence," from a group at Google Cloud AI Research, posted to arXiv on May twenty-fifth, twenty-twenty-six …”
Episode 089 — When AI-Written Papers Read Well But the Evidence Underneath Is Broken

Mentioned in 1 episode

  1. 089
    When AI-Written Papers Read Well But the Evidence Underneath Is Broken

Related terms