What would a JSONL schema for requests, limits and grants look like?

A workable layout uses one file per record family, joined by identifiers. Field names can be adapted in a pilot.

**`authority.jsonl`**

- `auth_id`, `role_id`, `limit_type` (single-item, aggregate, period) - `limit_value`, `currency`, `category`, `risk_band` - `effective_from`, `effective_to`

**`delegations.jsonl`**

- `del_id`, `grantor_role`, `grantee_role`, `scope` (amount cap, excluded categories) - `valid_from`, `valid_to`, `revoked_at` (nullable)

**`requests.jsonl`**

- `req_id`, `requester_role`, `assigned_role`, `amount`, `category`, `risk_flag` - `decision_date`, `linked_req_ids`, `evidence_refs`

**`routing.jsonl`**

- `event_id`, `req_id`, `actor_role`, `action`, `timestamp`, `comment`

**`gold.jsonl`**

- `req_id`, `permitted_outcomes`, `valid_approver_roles`, `condition_tags`, `rationale`

**Joins.** Resolve authority by matching `assigned_role` and `decision_date` against `authority` and `delegations`. Use `linked_req_ids` to compute aggregates.

**Formats.** JSONL suits streaming into a harness; the same tables can be exported as Parquet for analysis, with dates as typed columns.

The content is generated to this schema. Validate foreign keys and date logic on the sample, and check grounding and disclosure risk before wider use.