In what file formats and packaging can an intake release be delivered?

Resellers usually need a release to arrive in forms their customers can load without rework. A scoped intake release can be delivered in these shapes.

**JSONL event stream.** One line per event, ordered by `application_id` and `seq`, such as:

`{"application_id":"APP-2187","seq":4,"event":"check_result","rule_id":"R-INC-12","outcome":"fail","ts":"2026-03-02T09:14:00Z"}`

**Parquet tables.** One table per entity (`application`, `submission_item`, `check_result`, `routing_decision`, `handover`, `policy_rule`), with foreign keys on `application_id` and `rule_id`. Columnar format suits analytics and bulk loading.

**Task pack as JSONL.** Each task holds `task_id`, `initial_state_ref`, `tools`, `reference_path`, `scoring` and `reset_seed`.

**Manifest.** A file listing schema version, table names, key definitions, split assignments and generation parameters, so you can version your catalogue entry against it.

**Policy files.** Versioned text or JSON, referenced by `rule_id`.

For your own pipeline, ask for a schema file for each table, such as JSON Schema, so you can validate on ingest. Ask for stable column names across releases so downstream customers are not broken by renames. Where a schema changes, ask for a change note you can pass on to buyers.

Confirm exact formats in scoping. The list above is what you can ask for, not a fixed menu.