Phase 0 — execution status

Snapshot of the Phase-0 deliverables from plan/05-roadmap.md. All six in-flight sub-agents have reported back; the deliverables below are merged on this branch and ready for review.

Shipped

Component Path Verification
Application schema seed schema/deep/application.sql 40 nodes, 75 edges, 73 properties; loads + re-loads idempotently in DuckDB 1.5.2; kg.v_observation_summary view returns the expected wide row
knowledge-graph-mcp (read-only kg query MCP) mcp/knowledge-graph-mcp/ 12 MCP tools + 3 resources; 22/22 unit tests pass; smoke-tested against the live schema (572 nodes, 791 edges, 1027 properties loaded)
great-az-tick-check-mcp (mock submission tracking) mcp/great-az-tick-check-mcp/ 5 tools + 2 resources; 10/10 tests; mock backend with a clean override path for the real Walker-lab API
nws-heatrisk-mcp (NWS HeatRisk + heat-index) mcp/nws-heatrisk-mcp/ 7 tools + 2 resources; 17/17 heat-index tests against the canonical NWS table values; HeatRisk feed URL env-overridable
agents/ — 8-agent pipeline agents/ 9 agent modules + orchestrator; 29 unit tests; both Scenario A (tick mail-in) and Scenario C (CHW heat check-in) run end-to-end through the stub agents and assert the expected TriageDecision + Notification; failing-agent degradation test passes
app/ — Phase-0 UI prototype app/ 10 static files (vanilla HTML+JS+CSS, no build); the full 6-step tick mail-in flow renders cleanly; mock backend short-circuits to a canned response so the success card always renders

Sub-agent MCP tool inventory (cross-check)

Each agents/ module makes the following MCP calls — the corresponding mcp/<name>-mcp/ server should expose each:

Server Tool Caller
knowledge-graph-mcp ✅ regions_at_point GeoEnrichmentAgent
knowledge-graph-mcp ✅ outbreak_check EnrichmentAgent (VBD), KnowledgeUpdateAgent
vectorsurv-mcp ✅ agency_region_intersect GeoEnrichmentAgent
vectorsurv-mcp ✅ get_pools EnrichmentAgent (VBD), KnowledgeUpdateAgent
great-az-tick-check-mcp ✅ create_submission EnrichmentAgent (mail-to-walker-lab branch)
nws-heatrisk-mcp ✅ heatrisk EnrichmentAgent (Heat), KnowledgeUpdateAgent
mag-hrn-mcp ⏳ search_centers EnrichmentAgent (cooling center)
211-az-mcp ⏳ transport_to_cooling_center EnrichmentAgent (dispatch)

The two ⏳ servers fall through the orchestrator’s failure-isolation boundary today (observation still lands, with a failed_tools record). They’re queued for Phase 1.

Follow-ups flagged by the sub-agents

These are real but small; bundling them into the next PR rather than holding Phase 0:

  1. schema/deep/outbreaks.sql references pathogen.y_pestis but pathogens.sql defines pathogen.yersinia_pestis. Pick one slug and update the other.
  2. schema/deep/application.sql references code.snomed.heatstroke which isn’t defined in standards.sql. Add the SNOMED catalog row.
  3. schema/deep/counties.sql and tribes.sql lack lat / lon numeric properties, so knowledge-graph-mcp.kg_regions_at_point returns empty. Add a property-only follow-up seed (the tool already accepts lat / latitude / centroid_lat aliases).
  4. SNOMED catalog in standards.sql is heatstroke-only; the application schema would benefit from 84362002 (heat exhaustion), 52613005 (heat cramp), 24079001 (heat syncope) so the heat symptoms in application.sql chain through SNOMED via the existing crossReferences edges.
  5. LOINC code nodes for wearable observations (code.loinc.heart_rate_8867_4, etc.) so wearable_metric.* → mappedTo → LOINC works without storing the code as a string.

Plan-document gaps the sub-agents flagged

While building, the agents found six places where the plan was under-specified:

  1. Triage class list never enumerated in plan/03. Should reference schema/deep/application.sql or list the ten tc.* classes inline.
  2. Heat vulnerability score lacks a full factor table. Scenario C in plan/03 showed five line items; a full HEAT_SCORE_TABLE covering age 65+, thermo-meds, energy insecurity, and Red / Orange HeatRisk levels should be pinned in plan/01 or plan/03.
  3. consent.tick_mailin suppression trigger (“unless the submitter has been bitten”) isn’t formal. The agent treats “any Human field already populated → keep them, else zero out.”
  4. AgentRun audit-log schema mentions input, output, model id, latency, cost but cost / token counts aren’t on the model yet.
  5. Notification-ordering “life-threatening” set isn’t enumerated. tc.call_911 is the only entry today; should tc.urgent_care join it?
  6. Phase-1 MCP slugs (mag-hrn-mcp, adhs-mcp, whispers-mcp, inaturalist-mcp, 211-az-mcp) are referenced by string by the orchestrator but don’t have folders yet. They’ll wire on first creation.

What’s next

Phase 0 is functionally complete. Phase 1 (per 05-roadmap.md) adds the Heat vertical’s mobile flows, the Cluster Detection Agent, and three more MCP servers (mag-hrn-mcp, adhs-mcp, 211-az-mcp).