AIREP v0.2.0-alpha.1 (experimental prerelease)

Experimental prerelease. Not stable. v0.2.0-alpha.1 does not supersede AIREP v0.1 — v0.1 remains the recommended implementation target. There is no producer implementation yet, and the AD-15 stable-release criteria remain unmet. There is no stable v0.2.0.

The wire version remains 0.2 (airep_version is pinned "const": "0.2" in the shared schema). Release SemVer and the wire identifier are separate: an artifact says 0.2; the prerelease that shipped this tooling is v0.2.0-alpha.1.

Archived artifact

The citable, archived artifact for this prerelease is the Zenodo record, not the auto-generated source archives attached below:

• Version DOI (this prerelease): 10.5281/zenodo.22101986
• Concept DOI (all versions): 10.5281/zenodo.20475136
• v0.1 version DOI: 10.5281/zenodo.20475137 — cite this if you are building against AIREP today

GitHub generates its own source archives for every release; those are not the archived artifact and their bytes differ from the Zenodo upload. Use the Zenodo version DOI when you need an exact, fixed reference.

Protocol paper

AIREP: A Protocol for Per-Decision Evidence in AI Runtime Governance — arXiv:2608.21363 (preprint, not peer reviewed).

What this prerelease contains

• the frozen v0.2 integrity construction
• five JSON Schema files: one common definitions schema plus four artifact-family schemas
• a 117-fixture two-engine schema-validation corpus
• separately authored Python and Node class verifiers
• 60 scored class-verification cases plus 15 process/CLI probes
• a parity comparator covering G1–G14
• five negative proofs
• the preserved first failing parity measurement alongside the later passing evidence
• a vendored, hash-verified dependency basis for offline/no-index reproduction

Frozen identities at this tag

Artifact	SHA-256
CLASS_VERIFIER_CONTRACT.md	7ecfce56ab576a495816df77e25442b25c1afdb22cc9828e47ba29a565138885
verifier_py/class_verifier.py	5d08c327648d4bdc83714879be8531c837b991dd474d7ca46397b0ff8c9d01cc
verifier_node_r2/class_verifier.mjs	e678ff5706547d4fb79ab8ad013bdf6f41e4429065a42309d6a4a6515632bde4
corpus manifest aggregate	55f5189e38dfc81c6be68b5a51c789d5ed0f3ac8b0e57becfd7d796f1fddda4a

What the measured result does and does not establish

The class-verification result establishes agreement between the two implementations across the tested surface, and separately that each matches the frozen expected values on all 60 cases.

It does not establish third-party validation, third-party audit, correctness of the underlying semantics, truth or completeness of any real-world evidence, or stable AIREP v0.2 conformance. Agreement between the two implementations is consistent with separate authoring but is not proof of it.

The first official parity run failed, detecting two CLI/run-validity divergences. Its evidence is preserved alongside the later passing runs; the divergences were closed by clarifying the specification and correcting the diverging implementation, and the checks that detected them were not relaxed.