ARF v0.1

What ARF is

ARF is not a new format. It is the format Atlas already implements, written down so the same bytes and digests can be checked outside the product. If you already produce or store Atlas records, you are already on ARF; this document names the contracts.

Validation is fully offline: no network calls, and schema $id values are never dereferenced. Those identifiers (namespace path korvo.xyz/arf/ns/v0.1) name types — they are not fetch targets, and nothing serves them yet.

Record types

Ten record kinds make up the v0.1 vocabulary. Each has a JSON Schema in the Atlas repository.

• Question

  The query or prompt under evaluation.

• Source

  An external or attached input the evaluation may cite.

• Claim

  A discrete assertion produced or contested in the run.

• Artifact

  A content object (text, file ref, or structured payload) bound into a bundle.

• Validator

  How a check was applied and what it reported.

• Challenge

  A formal objection or counter to a claim or result.

• Endorsement

  Support for a claim, artifact, or evaluation outcome.

• Revision

  A superseding edit that preserves prior identity links.

• ChainRecord

  A link in an integrity or provenance chain.

• Evaluation

  The scored or judged outcome of a multi-model run.

Canonicalization

Bundles are reduced to a single digest so independent implementations can agree on the hash before any anchor or publication step.

Spec on GitHub

Authoritative text and schemas live in the Atlas repository under spec/v0.1/.

• spec/v0.1/README.mdOverview of ARF v0.1 and how the pieces fit together.
• canonicalization.mdRFC 8785 (JCS) rules and the sha256:<hex> digest format.
• disclosure.mdWhat is disclosed on the wire vs held at rest.
• bundle.schema.jsonJSON Schema for the ArtifactBundle envelope.
• examples/Worked examples for independent checking.