The Quilt That Checks Itself
Corroborated Discovery for the Internet of AI Agents — the core method and the atomic artifact, the Corroboration Record.
↗ PAPERThe Index Points, the Registry Proves
A quilt-of-registries model for verifiable agent discovery across NANDA and ANS — the ecosystem the bridge spans.
↗ INTERNET-DRAFTdraft-chandra-agent-registry-corroboration-01
Multi-Source Corroboration for AI Agent Discovery — the normative procedure, with sm-resolver and sm-divergence as its reference implementations. Current revision, submitted 27 August 2026.
↗ DATATRACKERRevision history
All revisions, status, and the diff between them. -00 was submitted 4 July 2026 and is superseded by -01.
Each revision is tagged to the code that implements it, so the normative text and the implementation move together: sm-divergence · draft-01 (sm-divergence 0.9.0 + the sm-resolver 0.3.0 kernel), and draft-00 (0.8.0 + 0.2.0) before it. The text committed at each tag is byte-identical to the revision the IETF serves.
What changed in -01
-01 adds the Answer Scope: a way for a source to declare, in advance, which
fields it varies and along which context. Not every difference between two registries is a
lie, and a procedure that reports every difference as one fires during normal operation.
The rest of the revision follows from it. A vantage becomes a controlled term with an
externally defined value space, so two sweepers mean the same thing by the same label.
Cross-source comparison is scoped to a cohort. A finding carries whether it was declared and
which revision of the declaration was matched. The record version becomes
corroboration/0.2, and a verifier is now required to check it.