What file layout and schemas could a booking release use?

One workable layout separates records, tasks and policy, and joins them through stable identifiers.

**Parquet tables** for record state: `approvals`, `facilities`, `accounts`, `schedules`, `collateral`, `fees`, `instructions`, `handoffs` and `exceptions`. Foreign keys are explicit, for example `accounts.facility_id` and `schedules.account_id`. Dates use ISO 8601, and amounts are stored as integer minor units with a currency column.

**JSONL tasks**, one per line:

- `task_id`, `instruction`, `start_state_ref` - `allowed_tools` and `difficulty_tags` - `expected_end_state_ref` pointing at the answer key

**JSONL answer keys** with expected record changes, acceptable alternative paths, required exceptions and handoff facts.

**Policy documents** as Markdown or plain text, with a table that maps each document to the product and rule IDs it defines.

**Environment definition:** a seed value, an initial state snapshot and a reset command that restores that snapshot, so the same task can run repeatedly.

**Manifest:** a file listing schema version, table row counts for that release, and how each file relates to the others.

Schemas can be adapted to your internal field names during scoping. Confirm the layout against a small sample before requesting more volume.