This view reads a privacy-safe ZCTA-week aggregation exported from the live DuckLake knowledge graph — the same store the intake write-path persists every report into. It is the analyst-facing read of the Cluster Detection Agent: it highlights where a category of report is arriving above its expected statewide share.
What this is — and is not.
Every signal below is an above-baseline count of what people reported
(e.g. febrile/rash reports, gastrointestinal/food-safety reports,
animal die-off/bite reports). It is never a diagnosis:
no disease, pathogen, or cause is inferred or named here, by design.
Confirming a cause is a downstream, human, agency act
(Scenario D),
not something this surveillance feed asserts. Aggregations are ZCTA-week;
no individual observation, precise location, or free-text note is exposed.
Provenance
Signals above expected statewide share · most recent window
ZCTA-week counts report-category counts by ZIP · weekly trend
| ZCTA | Place | Report category | Recent (2 wk) | Total to date | Weekly trend |
|---|
Open the underlying aggregation
Click to view the SQL behind this view. The dashboard never executes SQL
from the browser; analysts run it against the read-only
knowledge-graph-mcp endpoint configured for their agency.
Future work (not in this prototype):
live refresh on a schedule (this snapshot is exported by
scripts/export_signals.py); calibrated detection thresholds
per signal group; agency scoping so each audience sees only its
jurisdiction; and the audited "Verify" write-back through the
Knowledge Update Agent.
Cross-references: agency dashboard Arizona map pathogen graph field app