Crisis Routing¶
Safety preempts everything. Mira's crisis detection runs before consent checks, before bootstrap, before any other processing. A caregiver in distress gets help regardless of their onboarding status.
Gate ordering¶
This ordering was formalized in ADR-000112. Crisis detection (risk.level === 'high' || 'critical') preempts everything regardless of the caregiver's opt-in status, loop, or bootstrap stage.
Legal basis: TCPA emergency exemption (FCC Order DA 20-402; 47 C.F.R. § 64.1200(a)). Industry precedent: Crisis Text Line (741741) operates with zero consent gate.
Safety tiers¶
| Tier | Indicators | Response | Escalation |
|---|---|---|---|
| Distress | Overwhelm, high stress, thoughts of harm | Acknowledge + hotline (988) + follow-up | Human review queue |
| Imminent | Active suicidal/homicidal ideation, self-harm risk | Immediate crisis protocol + 988 + 741741 + 911 if physical danger | Mandatory follow-up, emergency contact if allowed |
| Continuity | Post-crisis monitoring | Structured check-ins at defined cadence | Step-down when stable |
Response format¶
First crisis response must include 988 (Suicide & Crisis Lifeline). Follow-up must include 741741 (Crisis Text Line) if not already included. Add 911 when there is immediate physical danger.
False positive recovery¶
- System: "I may have misunderstood. Are you doing okay?"
- If user confirms okay → safety flag resolves with reason
user_confirmed_safe
Two-step recovery prevents both missed crises (by asking) and stuck states (by accepting confirmation).
Exit criteria¶
- Distress: user sends non-crisis inbound → status moves to
monitoring→ clear when stable - Imminent: user confirms safety + 7 days elapsed →
resolved
Connection to zones¶
P6 (Emotional Wellbeing) is the primary zone feeding crisis detection. Spike detection in the GiveCare Score — a >= 30-point decline triggers immediate outreach and safety flag review.