Request for ARC question-to-exam/source provenance sidecar semantics

We are auditing a public research protocol that requires source-disjoint splits for ARC Challenge and ARC Easy. The official ARC archive exposes questionID plus CSV fields including examName, schoolGrade, and year, but the checked release does not document a complete per-row mapping to a genuine exam form or source document, and some candidate groupings remain ambiguous.

Is there an immutable public sidecar, schema note, or archived source table that defines the semantics of these fields and maps every released row to an exam form/document identity? If so, could you provide the exact release/commit, field semantics, license, file size and SHA-256, including duplicate and one-to-many-origin handling? Anonymized stable source-unit IDs are sufficient; we do not need question/answer text and will not treat filename, year, grade, or row identity as provenance without documentation.