Step 1 — The Data Source Cohort
Everything downstream is built from a cohort of addiction-treatment providers assembled from four directories plus the content we harvested from their websites. This page is the raw material.
The five sources
National roster of CARF-accredited providers. The only source carrying EINs — the bridge to IRS Form 990 financials.
Behavioral-health programs accredited by the Joint Commission, with location and service signals.
National Association of Addiction Treatment Providers — the highest-quality content cohort and our editorial quality bar.
Every facility carries latitude/longitude and a service category — the geographic spine for joins.
Full page/post content scraped from every provider domain exposing an open WP REST API, converted to a deduped Markdown corpus.
Why WordPress was the unlock
Reading thousands of provider sites by hand is a non-starter. But nearly half of them run WordPress with the REST API left open — meaning their entire catalog of pages and blog posts is available as structured JSON. That “fast lane” is what made a corpus of this size tractable.
We probed every domain, harvested the ~3,600 that answered, and converted 18 GB of raw HTML into a 4.6 GB Markdown corpus — word-counted, content-hashed, and deduplicated (11.7% of content was duplicated across the set).
- NAATP — 66% expose a WP API (best content, our quality bar)
- Joint Commission — 58%
- CARF — 43% (largest roster, lower WP rate)
The 4.6 GB corpus itself isn't shipped to this review site — it is represented downstream by the blog inventories and topic mine built from it.