docs: Update vision.md with full detailed content

docs/vision.md

@@ -96,4 +96,47 @@ Two important first-class concepts:
 2. **Contextual Telemetry Relationships**
    - Defined at alert creation time (with AI assistance)
    - Used to provide relevant surrounding data when an alert fires
-   - Static per alert rule
+   - Static per alert rule
+
+## 7. Device Provisioning & Connectivity
+
+- Every device includes cellular capability (baked into service tiers).
+- Primary connection method is cellular for most installations.
+- Zero-touch / low-touch provisioning is required.
+- Device uses a temporary hotspot during initial setup.
+- eSIM is under consideration for cleaner hardware and factory provisioning.
+- Devices are designed to be largely sealed after deployment (tamper-evident).
+
+## 8. Regulatory & Compliance Features
+
+- Planned as a **Premium tier** feature.
+- Goal is to reduce the administrative burden of regulatory compliance.
+- Includes tracking of testing schedules, generating reports, and maintaining audit-ready records.
+- Regulatory requirements are highly jurisdiction-specific, requiring a structured knowledge base.
+
+## 9. Technical Philosophy
+
+- Hybrid architecture: Local intelligence on the edge + stronger models in the cloud.
+- Strong preference for **Monitored Objects** and relationship-aware modeling.
+- Alert tuning is a core MVP capability.
+- Long-term vision includes a specialized fine-tuned model focused on water system monitoring and anomaly detection.
+
+## 10. MVP Scope Summary
+
+**In Scope for MVP**:
+- Basic and Advanced dashboard modes
+- Monitored Objects concept
+- Multi-tenancy (Owner + Operator)
+- Alert creation, management, and basic tuning
+- Contextual telemetry relationships
+- Simple RBAC with three roles
+- Operator Dashboard
+- Local device with cellular connectivity
+- Basic alert annotations
+
+**Out of Scope for MVP**:
+- Full Field Tech granular delegation
+- Advanced alert routing/suppression
+- Comprehensive regulatory automation
+- Needle gauge interpretation via computer vision
+- Custom complex rule builder (beyond basic threshold + dwell time tuning)