The data
Every fact points at a file you can download and hash yourself.
GetFunded is built only from public government records. This page lists the sources, shows what the database holds right now, says what is missing, and explains the rules for contact data.
Coverage
What the database holds.
Live from the database, cached for a few hours.
Organizations by type
2,309,073 in total. Database refreshed
| Type | Count |
|---|---|
| Public charities | 1,870,317 |
| Private funds (SEC Form D and ADV) | 180,174 |
| Private foundations | 166,036 |
| Companies | 68,890 |
| Investment advisers | 16,355 |
| Private equity advisers | 4,015 |
| Venture capital advisers | 3,268 |
| Federal agencies | 18 |
Do foundations accept applications?
What each private foundation says on its latest Form 990-PF.
| What the filing says | Foundations |
|---|---|
| Accepts applications | 28,238 |
| Funds preselected organizations only | 107,273 |
| Not stated in filings | 18,412 |
“Not stated in filings” is the absence of a statement, not a closed door. Public charities file a form with no application section, so they are not in this table at all.
Sources
Ten sources. All public.
URLs, refresh cadence and the limits of each source are in corpus/docs/DATA-SOURCES.md.
| Source | Publisher | What we take | Terms |
|---|---|---|---|
| Form 990 / 990-PF e-file index and XML | Internal Revenue Service | Filings spine; 990-PF officers, grants paid, financial lines, Schedule B, Part XV application info; Form 990 Schedule I grants, core financials, Part IX split, Part VII compensation; filer-stated websites | US government work, public domain |
| Exempt Organizations Business Master File | Internal Revenue Service | Identity: name, EIN, address, subsection, foundation code, NTEE code, ruling date, asset, income and revenue amounts | Public domain |
| Automatic Revocation of Exemption ListThe IRS dated the file we hold . | Internal Revenue Service | For each EIN: the revocation date, the date the IRS posted it, and the reinstatement date when there is one. It holds automatic revocations only: organizations that filed no annual return or notice for three years in a row | Public domain |
| Publication 78 dataThe IRS dated the file we hold . | Internal Revenue Service | For each EIN: the IRS deductibility codes, which say what class of organization it is for tax-deductible gifts | Public domain |
| Form ADV (IAPD) | US Securities and Exchange Commission | Adviser identity, CRD and SEC file numbers, offices, assets under management, private-fund schedules | Public domain |
| Form D | US Securities and Exchange Commission | Exempt-offering notices: issuer, offering amounts, related persons as filed | Public domain |
| SBIR / STTR awards | US Small Business Administration | Award records: agency, program, phase, amount, awardee | Public domain |
| ZCTA crosswalk | US Census Bureau | ZIP-to-ZCTA geography crosswalk used to normalize places | Public domain |
| Federal agencies and programs | This project | 10 agencies and 16 non-dilutive federal programs, curated by pull request | CC BY 4.0 |
| Entity-resolution labels | This project | Human match / not-match decisions that gate record linking | CC BY 4.0 |
Known limits
What is not here, said plainly.
A database that hides its gaps is not trustworthy. These are the ones we know about.
- The same fund can appear more than once. Records of one fund from different SEC sources are kept as separate organizations, and people are kept per source; we have not merged them.
- Not ingested: 990-EZ, 990-N, 990-T, paper returns and determination letters.
- IRS standing is only as current as the IRS lists we hold, and each statement shows the date of its list. The IRS replaces the lists about once a month. The revocation list holds automatic revocations only (no return or notice filed for three years in a row), so an organization that lost its status in another way is not on it and can still read as listed.
- The lines under “Can I apply?” that count a foundation’s past returns cover Form 990-PF returns the IRS has published as data. The count of recipients that are not on earlier grant lists compares names as written, and it can be wrong in both directions: a recipient whose name is written differently from year to year can look new, and two recipients with almost the same name in one state can be counted as one. It is shown only for a foundation with named grant rows four years in a row. It is a count from past returns, not a sign that a foundation will consider a new request.
- Some grant recipients are linked to an organization because three or more grant-making charities wrote the same name and state with one EIN on their own returns. No person has checked these links one by one. A link can point at a parent body, such as a parish or a sponsor, and the name on the row stays as the funder wrote it.
- The gives-per-year filter and the most-giving sort read the Form 990-PF giving lines (qualifying distributions, else charitable disbursements), so in name and browse searches they cover private foundations only; describe-the-work search also counts a public charity's reported grants, averaged per year. Profile grant lists include Schedule I grants from public charities as well as 990-PF grants.
- The IRS publishes XML in batches. At any time tens of thousands of indexed returns have no detail yet. They appear when the IRS publishes them.
- Back-year 990-PF grant rows are loaded for the newest index years only; earlier years carry filings, financials and officers.
- Form D names the issuer raising money, never the investors. It is not a deal graph.
- Inside parsed 990-PFs, the donor list (Schedule B) is present on about a quarter of filings, and the how-to-apply section (Part XV) says something usable on about a quarter.
- Keyword search matches names and titles only. Describe-the-work search reads giving records, so it covers private foundations, companies, federal programs and the public charities that report the grants they make; a charity with no grant list is found by name or EIN only.
- Filers make mistakes. We publish filings as filed. Out-of-range numbers are set to missing, not corrected.
The provenance contract
One chain, kept for every row.
Every fact on a funder page points back to the exact government file it came from, and to the exact record inside that file. If we read a file again later, the old facts cannot pretend to be new.
How it looks on a funder page (example values)
The dataset, the filing year, the fingerprint of the file we read (the first characters of its sha256 hash) and a link to the filing. When a fact has no file behind it, such as the identity line from the IRS master file, the fingerprint is left out rather than shown blank.
- Every file is kept and fingerprinted. We store the file as we downloaded it, its sha256 hash, when we fetched it, the publisher’s last-modified date, and when we last read it.
- Every fact knows its place in the file. A CSV row or an element inside the 990 XML, so anyone can open the same file and find the same value.
- Amended filings replace the original. Within one EIN, return type and tax period, the newest filing is the one that counts. Older ones keep their detail rows but lose their grant rows, so nothing is counted twice.
- Exports are reproducible. A published export is byte-identical on re-run and its manifest lists the hash of every file.
How it works in detail
dataset name → source URL → sha256-hashed immutable file → licence code
→ ingestion-ledger run → row (raw_file_id + source_record_locator)Record locators look like row:EIN=… for a CSV row or an element path inside the 990 XML for a filing value. Full description: PROVENANCE.md
Contact data
Published by affirmative act, never by default.
A funder page shows a contact only when the record allows it. Otherwise the database sends no value at all, so a bug in the app cannot leak it.
| Rule | What it means |
|---|---|
| Published on purposepublishability = public | Every contact record starts as internal only. Publishing it is a deliberate flag, set by the loader under one rule written once in the database, so every reviewer applies the same rule. |
| Role inboxes only | grants@ and info@ are desks and can publish. A named person’s address is withheld by policy, even when it appears on a public filing. |
| Three privacy tiers | Green is a shared desk such as grants@ or an office phone. Yellow is a professional contact for a named person, taken from a filing; it stays internal. Red is never shown. The database refuses to mark a red contact public. |
| No vendors, no scraping | Vendor contact data is never used. Facts from funder websites are never republished. A trigger refuses to mark either as public. |
| Checked before every export | Seven safety checks run before a single byte is written. If one fails, the export stops. |
License and corrections
Use it. Credit it. Tell us when it is wrong.
CC BY 4.0 for the compilation
Our selection, cleaning, linking and arrangement is CC BY 4.0. The government records underneath are public domain and need no credit. When you reuse the dataset, say:
Data from GetFunded, Open Funder Database (https://getfunded.ai), licensed CC BY 4.0. Derived from IRS, SEC, SBA, and US Census public records.
We publish filings as filed
Filings can be late, amended or wrong. We filter superseded filings; we do not correct the record. If a profile looks wrong, open a data correction issue. We trace the fact to its file and either fix our parser, load a newer filing, or leave it as filed with a note.
The dataset is provided as is, without warranty of any kind.
The full guide: Data sources and license.
See the data for yourself.
Search is free and needs no account. Every fact on a funder page carries its source line.