Methodology
Where the data comes from.
Every number on Fonte House is public-record, cited, and refreshed on a schedule. Three sources, three cadences, one workspace.
01 IRS 990-PF filings
Every private foundation across Rhode Island, Massachusetts, Connecticut, and New York that files IRS Form 990-PF is in the index. We pull the full filing, officers, compensation, schedule of contributions, grants disbursed, from the IRS Tax Exempt Organization Search bulk archive.
Coverage as of today: 17,997 funders · 644,071 grants on file. Updated nightly at 06:00 UTC; any new filings the IRS publishes appear within 24 hours.
Public-by-statute under 26 U.S.C. § 6104. Officer compensation and grant amounts are reported directly from the form as filed.
02 Donor recognition
We scrape nonprofit donor-wall pages across Rhode Island, Massachusetts, Connecticut, and New York: universities, hospitals, arts organizations, civic groups, and foundations. Names, giving tier ranges (not exact amounts), and the recognition year when available. Public-by-publication.
Coverage: 668 recognition pages across the four-state index. Recognition data is what makes cross-organization giving patterns visible, the core of the Match feature.
Tier ranges (e.g. "$10K-$25K") are inferred from the page section where the donor appears. We do not publish exact dollar amounts because the source nonprofits don't.
03 News mentions
Bing and Google News RSS, keyed on each funder's legal name. Catches new initiatives, controversies, leadership changes, and capital announcements. We dedupe across runs so the same story doesn't appear twice.
Free, rate-limit-polite. No API key required. Stored with the
unique constraint (funder_ein, url).
04 LLM categorization
Each grant where the IRS didn't supply an NTEE major category gets a category assigned by Claude Sonnet. Education, Health, Arts & Culture, and the other 22 NTEE letters. We use the LLM only when the IRS BMF lookup doesn't produce a match.
Transparent and inspectable. Every assignment is in the
grants.ntee_code_inferred column; the daily cron
preserves these on rebuild.
05 Privacy and ethics
Fonte House uses public-record data only. We do not process:
- Social Security numbers, financial account numbers, government IDs
- Health information, religious affiliation, political party membership
- Anything not already publicly available on the source site
Subject rights endpoints at /api/subject-rights/*
provide lookup, export, deletion, and correction for any individual
in the recognition index. Standard response time: 10 business days.
06 Sub-processors
Fonte House runs on Cloudflare (hosting, D1 database, edge compute), Anthropic (LLM categorization), and Resend (transactional email). All US-based, contracted under standard DPA terms.
Data last refreshed: checking...