# Occupation AI Workflow Route Card

Use this card before applying an AI workflow to a job.

```yaml
occupation_and_role:
employer_or_business_context:
jurisdiction_or_program:
license_registration_or_authority:
job_to_be_done:
people_and_rights_affected:
money_or_property_affected:
private_or_restricted_data:
approved_systems:
exact_ai_preparation_job:
human_decisions_and_actions:
qualified_reviewer:
proof_required:
terminal_state: approval_ready
```

## Route rules

1. Name one occupation and one task—not “automate the department.”
2. Choose a substantial specialist manual when the task is regulated, sensitive, safety-related, or consequential.
3. Use only the current authority sources that apply to the role, jurisdiction, product, program, organization, and facts.
4. Keep original inputs and evidence states separate from AI summaries and recommendations.
5. Let AI prepare a checklist, evidence ledger, draft, QA report, or handoff packet.
6. Keep licensed/professional judgment, rights, safety, money, messages, account writes, and public actions with the authorized human.
7. Bind approval to the exact payload, scope, recipient, amount/action, source version, approver, expiry, rollback, and receipt.

## Shared hard stops

- no fabricated authority, expertise, source, document, execution, result, review, or proof;
- no private or restricted data in unapproved AI systems;
- no hidden send, publish, deploy, file, purchase, payment, account, CRM, MLS, portal, production, or system-write action;
- no generic model output presented as licensed, regulated, clinical, legal, financial, safety, or technical judgment;
- no shallow occupation shell presented as a completed specialist guide.

Terminal state: `approval_ready`.