Entities and relationships
| Entity | Key relationships |
|---|---|
| Farmer | consents, farms, cases |
| Farm/Field | geometry, context cache, crop cycles |
| Case | turns, live sessions, evidence, assessments, escalations |
| LiveSession | room, participants, network mode, tracks, selected frames |
| Participant | farmer, AI agent, officer/agronomist |
| MediaTrack | audio/video metadata, start/end, quality telemetry; not necessarily recorded |
| FrameSample | timestamp, quality, objective, hash, storage reference and evidence ID |
| Turn | original/normalized transcript, entities, speaker |
| EvidenceItem | frame/API/farmer/document/research provenance |
| Assessment | hypotheses, evidence, policy results and response |
| Escalation | join request, reviewer action, SLA and disposition |
Snapshots
Every reasoning run reads an immutable case snapshot containing transcript turns, accepted frame IDs, context, evidence and all model/prompt/tool versions. Room telemetry may be aggregated, but any data that materially influenced a decision must be referenced.
Geospatial sensitivity
Separate precise field geometry from coarse district context. Grant precise access only when operationally required and record every access/export.