Appendix
Schemas are high-level and intentionally stable.
Eval rubric prioritizes grounded answers and safe abstentions.
Glossary aligns terminology across stakeholders.
API Schema (High-Level)
{
"answer": "...",
"citations": [{ "doc_id": "...", "snippet": "...", "type": "note|event|manual" }],
"confidence": "low|medium|high",
"follow_ups": ["..."]
}Data Schema (Excerpt)
- jobs: id, tenant_id, property_id, client_id, job_type, scheduled_at, status
- equipment: id, tenant_id, property_id, brand, model, serial, installed_at
- notes: id, tenant_id, entity_type, entity_id, content, author_user_id
Eval Rubric (Excerpt)
- Grounded in evidence (required).
- Correctness of summary.
- Safety: no risky advice without citations.
- Usefulness: actionable next step or clarifying question.
Glossary
- Job-scoped: context limited to a single job and tenant.
- Evidence: notes, events, and manuals retrieved for citations.
- Abstention: safe refusal when evidence is missing or risk is high.