Fair lending attestation
your examiner can verify
CREST delivers the HMDA reporting analytics institutions already file — plus a hash-anchored, examiner-ready attestation record they can defend. STRATA, our race-inference engine, replaces legacy BISG and cuts the most consequential error — White misclassification — by 37%.
arXiv 2504.21259 · 2025 LSTM+Geo with XGBoost Filtering: A Novel Approach for Race and Ethnicity Imputation with Reduced Bias arXiv 2505.16946 · 2025 STRATA: A Name-and-Geography Race Inference Model for Residential Deed Records SSRN · June 2026 Who Is Buying New York City's Small Homes?
on matched holdout
(28.5% → 18.0% vs BISG)
filed
live today
no data egress
attestation record
A better proxy alone isn't enough
BISG is the race-inference method every regulated lender uses for fair lending analysis. It systematically misclassifies minority borrowers — producing a 28.5% false positive rate on White classification. That's the specific failure mode examiners care about most: it causes institutions to systematically undercount disparate impact on minority borrowers.
But a better proxy alone is not enough. When any error propagates through a disparity analysis, the institution still can't tell whether the gap it measured is real or a measurement artifact. CREST answers that question — BISG has no answer at all.
| Model | White FPR |
|---|---|
| BISG (industry standard) | |
| STRATA aequumAI |
Source: STRATA M8 (name + 13 Census geo features) benchmarked against BISG, A&B, and ZRP on matched voter holdout n=986,801 and national PPP holdout n=125,081.
CREST produces the record.
STRATA powers the estimate.
Together they replace a black-box proxy workflow with a defensible reporting layer institutions can actually use — without replacing existing infrastructure.
Input
Institution uploads or batches borrower records through a stateless AWS or on-prem workflow. If race is self-reported, we use it directly. If not, STRATA supplies the inference path.
STRATA
Replace legacy BISG with a more accurate race-inference engine: 88.3% accuracy and a 37% reduction in White misclassification FPR (28.5% → 18.0%).
CREST
Simulate how residual classifier error propagates through the disparity analysis. Quantify the confidence band around the measured gap so the institution knows whether a result is real or a measurement artifact.
Output
Deliver HMDA reporting analytics formatted for direct filing use, plus a hash-anchored attestation record an examiner can receive, verify, and cite.
CREST — HMDA Filing & Fair Lending Attestation
- Loan file in via batch or API
- STRATA inference where race is missing
- Disparity analytics by group and geography
- Confidence-bounded output for filing use
- Hash-anchored, examiner-ready attestation
- Runs on AWS SageMaker or customer-controlled AWS deployment; no client data enters aequumAI infrastructure
Required reporting delivered with usable analytics
DELTA — Deposit-to-Lending Community Analytics
- Depositor and borrower records in
- Community benchmark at tract or zip level
- Deposit composition profile
- Lending profile
- Differential profile between depositors and borrowers
- Optional stored simulations for geolocated disparity graphics
Turns compliance into a publishable community metric
Deploy in your environment
CREST and DELTA run as a stateless workflow on your own customer-controlled AWS deployment. Borrower and depositor records never enter aequumAI infrastructure — which is exactly what your IT and legal teams need.
Five reasons regulated institutions pay for this today
ECOA, HMDA, and CRA compliance is not discretionary — and no one else offers auditable error quantification in a form examiners can use.
Legal Obligation
ECOA, HMDA, and CRA compliance is not discretionary. Every institution must analyze for disparities, and CREST is built to make that work auditable.
Examination Defense
When an examiner asks whether a measured disparity is real, CREST provides a reproducible, verifiable answer. That answer is worth more than the tool's cost.
Data Sovereignty
Customer-controlled AWS deployment means no client data enters aequumAI infrastructure. Borrower records stay inside the institution's own AWS environment at every stage — important for GLBA, Regulation B, and internal risk controls.
Enforcement Urgency
DOJ and CFPB fair lending enforcement actions have increased. Institutions that cannot defend their measurement methodology face examination findings, consent orders, and CRA downgrades.
Community Contribution Story
DELTA produces a deposit-to-lending differential profile showing whether deposits from one demographic group are being converted into loans serving the same community — turning CRA obligation into a publishable community contribution metric.
No one else quantifies the error
| Capability | CREST (aequumAI) | Compliance Consultants | Model Risk Vendors | DIY / Internal |
|---|---|---|---|---|
| Quantifies classifier error propagation | Yes | No | No | No |
| Examination-ready attestation record | Yes | Partial | No | No |
| On-premise, zero data egress | Yes | No | Partial | Yes |
| Patented methodology | Yes | No | No | No |
| Works with existing models | Yes | Yes | Partial | Yes |
Source: Company analysis of current alternatives, June 2026
First revenue signed. Federal methodology validated. State regulator engaged.
Three papers published, a fourth submitted to DMLR. STRATA and CREST are deployed on SageMaker and live on AWS Marketplace, enterprise procurement ready.
EQUITYchain™ trademark published May 5, 2026 in the USPTO registry — Serial No. 99486925, patent pending, pilot planned for FY2028+.
NY Hotel Trades Council
August 2026 delivery, with expansion in negotiation.
$1.6M, Zero Outside Capital
Four years of development ahead of this raise.
6 Filed
CREST core · CREST-2 · CREST-Optimal · CREST-Decompose · DELTA — pending provisional June 2026.
AWS Marketplace
STRATA + CREST deployed on SageMaker, enterprise procurement ready.
Built on four years of SSA research
aequumAI was founded by the team that developed race-modeling methodology at the Social Security Administration, producing multiple peer-reviewed research papers on demographic estimation. That research foundation — combined with comprehensive New York State real estate, voter, and property records — became the basis for STRATA and CREST.
Our published model (arXiv:2505.16946) demonstrates that STRATA's LSTM+Geo + XGBoost ensemble outperforms BISG on every measured dimension. CREST's error-propagation simulation methodology is unpublished, patent-pending, and extends the published work significantly.
STRATA: Published LSTM+Geo Model
Neural network combining name-based surname probability with geolocation. Outperforms BISG and BIFSG on accuracy and White false positive rate. Extended to 6 race categories including Native American.
CREST: Spatially Correlated Error Simulation
Errors within a geography are modeled as correlated, not independent. Produces fat-tailed uncertainty distributions that correctly estimate the probability of systematic misclassification across a population segment.
Ground Truth Reliability Adjustment
In certain geographies, self-reported race is systematically corrupted by non-response bias — as documented in PPP Hispanic underreporting in AZ, NV, and UT. Our model adjusts for ground truth corruption rather than treating divergence as model error.
Enterprise AWS Deployment
Stateless CREST and DELTA workflows deployable on customer-controlled AWS or via AWS Marketplace. No data egress. Annual license with methodology updates.
Who Is Buying New York City's Small Homes?
A decomposition of 2.26 million NYC 1–3 family deed transactions (1966–2025), with buyer and seller race imputed by STRATA scored on each party's own mailing address — not the property's neighborhood demographics.
The Trust Surge
- The standard "LLC/Inc/Corp/Trust" flag reports 45.0% corporate share of 2025 NYC purchases
- Decomposed: 23.9% genuine investor capture, 21.1% family trusts
- Family trusts are overwhelmingly families holding their own homes (84% own-race rate vs. 29% baseline)
A Crisis Step, Not a Climb
- Genuine investor share rose from ~10% pre-crisis to ~24% by 2025
- A one-time structural step at the 2008 financial crisis, plateaued at ~2.5× baseline for a decade
- Not an accelerating takeover — policy should target the plateau, not a crisis
The LLC-Transparency Gap
- Net Black homeownership growth collapsed from +3,400/yr in 2008 to +90/yr in 2025
- Driven by sales to investor entities whose beneficial owner race is unobservable — not white in-migration, not the trust surge
- Beneficial-ownership disclosure is the binding constraint on evidence-based housing policy
Chalavadi, Leitch & Pastor · SSRN Working Paper, aequumAI LLC · June 2026
Home Ownership by Race in New York State
Using NY State and City tax records — name, address, and assessed value — we estimated racial homeownership composition across census tracts and compared predicted ownership percentages against Census-reported demographics.
White & Hispanic Ownership — Income Analysis
- White individuals only: ~85.30% ownership. Overrepresented in most low-income tracts, near parity in higher-income areas.
- With corporate included: ~77.60%. Overrepresentation decreases but remains.
- Hispanic individuals only: ~5.29%. Underrepresented across the income spectrum.
- With corporate included: ~4.81%. Underrepresentation deepens in lower-income areas.
White & African American Ownership — Racial Analysis
- White individuals only: ~85.29% total. Generally overrepresented.
- Including corporate: Drops to ~77.55%, still above parity in many tracts.
- Black / African American individuals only: ~2.58% total. Mostly underrepresented.
- With corporate: Further decreases to ~2.35%, deepening the gap.
The people behind the model
Terry Leitch
25 years quantitative finance — Boeing, AEGON, BofA, ING Barings. CFTC consultant on ~$7T of OTC derivatives. SSA data science contract, ~$250K/yr. Columbia SPS faculty; 8 years at Johns Hopkins. BA Math (Chicago), MS Applied Math (Northwestern), MBA (London Business School). 3 arXiv papers, DMLR submission, 6 provisional patents.
Sanjana Chalavadi
MS Data Science & AI, University at Buffalo. Manages STRATA and CREST on SageMaker. Co-author, arXiv:2505.16946. Designed and delivered the 43North demo. STRATA is live on AWS Marketplace.
Andrei Pastor
PhD Physics, Florida State University. MS Financial Mathematics, Carnegie Mellon. Architected all production database systems and built the NYC deed pipeline (2.26M records). Co-author on race-proxy methods papers. Background in mortgage trading and structured products.
Ready to talk?
Whether you're a compliance team evaluating BISG alternatives, a state regulator, a channel partner, or a researcher — we'd like to hear from you.