Data

Four source families feed every model. Labeling and coverage are shared method; joins remain Data rules. Start with Framework · Labels and observability.

1 First American — Silver

Dated evidence Silver AE QA — 28 × 8 counties (2026-06-04) ↗ · #104 · ARS↔Smarty 91/104 (2026-06-15) ↗

FA data dictionaries33 sources, field-level Data sourcesproperty · owner · history FA roof codesROOF_COVER + assessor codes

2 BuildZoom — Gold

Dated evidence Gold-QA — 23 × 13 counties (run 2026-06-04; data vintage 2026-05-12) ↗ · #100 metadata; evidence is repo-only

Gold variable inventorypermit history, as landed BuildZoom dictionary46 delivered fields · source updated 2025-01-27 ETL contractsLanding → Bronze → Silver → Gold

3External public sources

Status · 2026-08-19planned Externals-QA — required per P1

Demographics catalogACS · Decennial · IRS · HUD · BLS · FCC US demographics cachenationwide parquet, all models Scraping catalogcounty portals, Stage 1

48020REI clients

Status · 2026-08-19in progress REI / Apollo research step 1.1 — join keys · date grain · deal volume; freezes the label spec

REI / Apollo label familysale event; dealable outcome is separate research Client data inputssources, fields and joins

Cross-source

Feature taxonomyvariables → features + synthetic families