What file layout and schema could a generated booking dataset use?

A release could be delivered as one file per entity, in JSONL for readability or Parquet for scale, with a manifest describing each file.

**Core tables** - `approvals`: `approval_id`, `applicant_id`, `product_code`, `approved_limit`, `pricing_tier`, `authority_role`, `approved_at`. - `conditions`: `condition_id`, `approval_id`, `text`, `status`, `evidence_ref`. - `facilities`: `facility_id`, `approval_id`, `limit`, `term_months`, `repayment_frequency`, `balloon_amount`, `status`. - `schedule_lines`: `facility_id`, `line_no`, `due_date`, `principal`, `interest`, `line_type`. - `security_items`: `security_id`, `facility_id`, `type`, `registered`. - `status_events`: `event_id`, `facility_id`, `from_status`, `to_status`, `actor_role`, `tool_action`, `reason`.

**Task files** - `tasks.jsonl`: `task_id`, `approval_id`, `start_snapshot`, `expected_snapshot`, `expected_disposition`, `blocking_items`, `difficulty_tags`.

**Manifest** Lists each file, primary and foreign keys, row counts for that delivery, schema version and generation seed.

Dates should be ISO 8601 and money fields should be integers in minor units or decimal strings, not floats, so schedule reconciliation is exact. Confirm the format and key conventions in a sample before scoping a full release.