How are intake records linked so a case can be replayed step by step?

Intake data is easiest to resell when its links are explicit. A release can model one case as a small graph of records joined by stable identifiers.

Core entities and keys:

- `application`: `application_id` (such as `APP-2187`), `channel`, `received_at`, `domain`, `current_state` - `submission_item`: `item_id`, `application_id`, `item_type`, `legibility`, `present` - `check_result`: `check_id`, `application_id`, `item_id`, `rule_id`, `outcome` (`pass`, `fail`, `waived`) - `routing_decision`: `decision_id`, `application_id`, `queue`, `reason_code`, `rule_id`, `decided_by_role` - `handover`: `handover_id`, `application_id`, `from_role`, `to_role`, `note` - `policy_rule`: `rule_id`, `version`, `text`, `effective_from`

Every event carries `application_id` and an increasing `seq`, so a consumer can replay a case to any point. The `rule_id` links checks and decisions to the policy file, which lets a buyer test whether an agent's output matches the stated rules.

For resellers, the useful property is that these keys survive repackaging. If you cut a release into customer-specific subsets, keep case-level integrity: split by `application_id`, never by row, so events from one case do not scatter across training and test sets. State that rule in your catalogue entry so buyers do not undo it.