Built in the open
Fundprint is a work in progress, maintained transparently. This is what has shipped, what is being measured, and what is next.
- Foundation
The five-stage pipeline
A deterministic Acquire, Store, Resolve, Validate, Publish architecture on PostgreSQL with pgvector. The contract between layers: shape, trust, and provenance.
- First registry
NPPES ingest and brand matching
The federal provider registry, read into a staging table, with deterministic brand-prefix matching of clinics to owner entities and owners to parent firms.
- Taxonomy
Honest owner types
Private equity, pension fund, family office, and other institutional owners labeled distinctly rather than lumped together as PE.
- Second registry
Owner location directories
A supplement to NPPES: reading each owner's own public directory from machine-readable structured data (schema.org and semantic address fields), de-duplicated against the registry.
- Validation
Confidence floors and a hand-validation gate
A 95% hand-validation gate, confidence floors per link type, quarantine for contradicted claims, and a versioned methodology behind every figure.
- Launch
The public record goes live
whofundsmytherapist.com ships as a pinned static snapshot: no runtime database, every claim traced to a content-hashed source.
- Platform
The Dossier: redesign, map, and the engine
A full investigative platform: an interactive national clinic map, the pipeline rendered as a live diagram, and the story told end to end.
- The market
A denominator, and one clinic per address
The whole national ABA market measured from the registry: 17,569 providers, 21,088 locations, and the size of every operator in it. Along the way, addresses where one owner had registered two of its brands were found to be one clinic each, not two, and were merged.
- Correction
A published share, withdrawn
Fundprint used to headline private equity's share of clinics run by operators with five or more locations. It was withdrawn: the cutoff was arbitrary, and the group it measured against is one private equity itself builds, so its own buying inflated both halves of the ratio. The headline is now a count, and the shares that remain need no cutoff at all. Recorded in the changelog rather than quietly dropped.
- Sources
The operator's own list beats the registry
A provider registration is filed once and never revoked, so the registry keeps a clinic on the books long after it closes and cannot see a centre a chain never registered separately. Where an operator publishes its own complete list of centres, that list now decides. It corrects in both directions: Autism Learning Partners rises from 1 clinic to 45, while Hopebridge falls from 145 to the 101 it actually lists, and Acorn Health from 90 to 70. The registrations left behind are held in quarantine, not deleted.
- Denominator
Coverage stated as a fraction, against someone else's list
A clinic count with no denominator invites the question 'out of how many?' and has no answer. Coverage is now stated as a fraction of the known private-equity-backed ABA platforms, measured against the appendix of the Private Equity Stakeholder Project's April 2026 report rather than a list of our own drawing. It runs both ways: that appendix omits four platforms published here, and it names eight more that are not, holding several hundred facilities between them. Every one of them is listed by name, with the reason.
- Reconciliation
The gap against the published estimate, taken apart
The peer-reviewed count for the same country is 574 private-equity-owned sites; this dataset holds 1,621. Rather than assert one is right, the difference is now decomposed and published. Both find private equity in exactly 42 states, so the disagreement is depth rather than footprint. Restricted to what the federal registry can see, the two counts land six sites apart, and the whole of the remainder is centres that appear only in an operator's own directory. The registry's blind spot runs from 1.0x to unbounded operator by operator, so no multiplier can correct for it.
- States
What the auditors found, next to who owns the clinics
Care is bought locally and so is oversight. Each audited state now has its own page pairing the government audit of its Medicaid autism-therapy spending with the ownership traced there. The pages refuse the obvious insinuation: no audit attributes a dollar to private equity, and Maine is published as the control case, with the third-largest finding of the four and no private-equity clinics at all. What the pairing does support is that Wisconsin's auditors can see 2 of the 48 private-equity-owned centres in their state.
- Next
A verification rate with its confidence interval
A stratified hand-check of published clinics against independent sources, reported as accuracy with a 95% confidence interval broken out by source type, plus the rate of registrations that survive after a centre has closed. Stated as a number, not implied.
The commit history is public across all three repositories, linked from the investigator.