Ingestion & Provenance

Every observation is sourced, dated, licensed, and confidenced.

A 15-stage pipeline turns raw records from every jurisdiction into evidenced graph facts — with conflicts preserved and reviewable.

15-stage staged pipeline

Source Registration

Stage 1/15

Raw Data Import

Stage 2/15

Schema Validation

Stage 3/15

Data Normalization

Stage 4/15

Address Standardization

Stage 5/15

Geospatial Validation

Stage 6/15

Identity Resolution

Stage 7/15

Duplicate Detection

Stage 8/15

Relationship Generation

Stage 9/15

Conflict Detection

Stage 10/15

Confidence Assignment

Stage 11/15

Property Graph Upsert

Stage 12/15

Score Recalculation

Stage 13/15

Audit Logging

Stage 14/15

Review Queue Creation

Stage 15/15

Connector framework

Data connectors — status is honest

County Parcel Data

Bulk + delta feeds per county

DEFERRED

Assessor Data

Assessment history + tax status

DEFERRED

Deed Records

Recorded transfers + encumbrances

DEFERRED

Tax Records

Delinquency + tax status

DEFERRED

Zoning Data

District boundaries + rule text

DEFERRED

Permit Data

Applications, filings, inspections

DEFERRED

Census Data

Geographies + demographics

STRUCTURAL

FEMA

Flood zones + risk

STRUCTURAL

Environmental Data

Contamination, wetlands, habitat

DEFERRED

Transportation

Transit, road hierarchy, ADT

STRUCTURAL

Utilities

Water, sewer, electric, gas territories

DEFERRED

Telecom

Fiber routes, macro sites

DEFERRED

Satellite Imagery

Sentinel-2 derivations

STRUCTURAL

Market Transactions

Sales comps + capital markets

DEFERRED

Commercial Data Providers

Requires license + contract

BLOCKED

Live national feeds require credentials and licensing. Mock adapters exist for UI testing only and are always labeled.

Source classifications

GovernmentCounty AssessorRecorder of DeedsMunicipalityGIS ProviderUtilityCommercial Data ProviderUser SubmittedAnalyst VerifiedAI DerivedCalculatedPublic RecordSatellite or Aerial Imagery

Every observation stores

· Source + source record ID
· Retrieval date + effective date
· License restrictions
· Confidence + verification status
· Transformation history
· Original value + normalized value

Conflict resolution — versioned, never overwritten

Field	Source A	Source B	Confidence	Status
properties.year_built	1968 (Assessor)	1969 (Planning)	0.98 / 0.72	Analyst Review Required
parcels.lot_area_sqft	22,400 (Assessor)	22,412 (Survey)	0.94 / 0.99	Automatically Resolved → 22,412
ownership.percent	100% (ACRIS)	50/50 (User Submitted)	0.99 / 0.42	Source Correction Pending
buildings.height_ft	420 (DOB)	418 (LiDAR)	0.98 / 0.86	Automatically Resolved → 420

Corrections generate corrective events. Historical values are never destroyed.

Migration & validation posture

Additive only

Additive migrations, no destructive renames, no dropped working tables.

Tested boundaries

Tenant isolation, RLS, duplicate handling, and score versioning tests.

Immutable ledger

Property events are append-only. Corrections create new corrective events.