Multi-Source Corroboration for AI Agent Discovery (draft-chandra-agent-registry-corroboration-02)

Abstract

AI agent discovery depends on sources such as registries, name services, DID methods, and catalogs. A valid signature on an artifact does not by itself show that a source has not withheld a record, or served inconsistent records to different observers. This document specifies a source- and format-agnostic procedure for corroborating independent observations of an agent. It defines claim classification, canonical views, deterministic disagreement findings, multi-vantage equivocation detection, propagation-delay confirmation, and signed Corroboration Records. The procedure produces verifiable evidence for every sweep, including agreement and insufficient- observation results. It composes with registry-assisted resolution of known subjects: a registry's renewable subject-to-target binding carries a sequence, and the procedure attributes a source serving a superseded binding rather than comparing it, so propagation lag is never reported as disagreement.