What integration contract should a lab agree for its first generated release?

Ask for a release manifest that identifies `release_id`, `schema_version`, `content_digest`, `licence_reference` and the included artefacts. Each task should carry a stable `task_id` and references to the exact generated records it requires. A lab can store these identifiers with experiment outputs to establish which material produced a result.

If both downloads and MCP access are in scope, agree how they relate. A paginated tool response should identify its release, and an equivalent export should preserve the same record identifiers. Fetching successive pages must not silently mix different revisions of a review history.

Specify how schema changes are communicated, including renamed fields, altered enums and changed null meanings. A small ingestion fixture can verify that the buyer's loader preserves relationships and rejects unsupported versions before a larger import begins.

MirrorFirm could scope JSONL, Parquet or tool-access delivery around the lab's existing pipeline. The useful first integration step is an agreed manifest and a successfully loaded fixture, with explicit handling for missing artefacts, duplicate records and release-version mismatches.