What technical handover should be agreed before connecting a generated release to an agent?

Ask for a release manifest that identifies schema versions, included files, record types and declared relationships. Field definitions should explain null values, enumerated states, date semantics and whether identifiers remain stable across revisions.

For file delivery, agree how JSONL or Parquet represents nested evidence references and optional attributes. A missing value should not ambiguously mean unavailable, not applicable and not yet assessed. These distinctions can change an agent's decision even when ingestion succeeds.

Map the proposed tool interface to your agent's adapter layer. Document argument names, result envelopes and error handling expectations before connecting the complete workflow. The first integration step can be a contract check that loads a supplied review, resolves its references and validates one permitted action against the declared schema.

Also ask how schema changes would be described and how a buyer could identify which release produced a stored run. Avoid assuming compatibility between revisions.

This handover establishes technical fit. It does not by itself establish business fidelity, so the loaded records should still receive a separate review against the pilot's intended operating rules.