Methodology

Every figure on this site is derived from a public IRS filing by a pipeline whose source is open. This page explains the derivation in the terms a grant consultant uses, so that a number here can be cited with the right amount of confidence and no more.

Where the rows come from

The IRS publishes every electronically filed Form 990-series return as XML. Private foundations list the grants they paid, and those approved for future payment, on Form 990-PF, Part XV. Public charities that make grants list them on Form 990, Schedule I, Part II. The pipeline reads every such row from the current IRS posting, maps each schema version's element names through the Nonprofit Open Data Collective's IRS E-file Master Concordance File, and writes one row per grant. Nothing is keyed by hand and nothing is rewritten: names, amounts and purposes are as filed.

Filers whose total paid is stated but whose grantee list is an attachment the XML does not carry are recorded as such and excluded from the edge list, with the stated total published separately. A funder absent from this site is not a funder that made no grants.

Two kinds of amounts, never added together

Form 990-PF reports grants paid during the year and grants approved for future payment in two separate tables, and they overlap across years: a grant approved in one year appears again as paid in the next. Summing both overstates giving, sometimes badly. This site keeps them apart on every page. Totals are grants paid; approved-for-future commitments are shown in their own card and their own column, and never enter a total.

How recipients are matched, and what the tiers mean

Form 990-PF usually names a recipient by name and address only. Turning that into an organization means matching it against the IRS Exempt Organizations Business Master File, and matching is inference. Every edge carries a tier, and the tier is the honest answer to "how do you know it is them?":

• A — Reported EIN. The filer reported this recipient's EIN on the return. Where the EIN is in the IRS Business Master File the match is verified; where it is not, the EIN is carried as reported.
• B — Exact name and place. The recipient's name, normalized, and its ZIP code or state match exactly one organization in the Business Master File. Deterministic, no scoring.
• C — Strong name match. The name is a close but inexact match (Jaro-Winkler at or above 0.94 on the normalized name) to one organization in the same state, with no comparable alternative.
• D — Probable name match. The name is a probable match (at or above 0.90) to one organization in the same state, with no comparable alternative. Treat as a lead, not a fact.
• U — Unresolved. No organization in the Business Master File could be identified with the required confidence. The grant is real; the recipient is shown as filed.

The matcher blocks candidates by exact normalized name, by state and first name token, by ZIP code and first token, and by state and phonetic key; scores the survivors on name similarity with small adjustments for a matching city and ZIP; and refuses any tuple with two candidates within a few points of each other, which is how chapter-style names ("Rotary Club of …", "American Legion Post …") stay unresolved rather than wrongly resolved. Grants to named individuals are excluded before any of this.

How the matching is verified

Precision is measured, not asserted, against a hand-labeled set of recipient tuples drawn evenly from every tier: at least a thousand pairs, each verified against the Business Master File and the filing by a person. The targets are 99% for tier A, 97% for tier B, 95% for tier C and 90% for tier D. A build that misses a target for a tier does not lower the target; it reports the miss. Until the check for a dataset version is complete and published here, the site says so on every entity page, and tiers B–D should be read as leads.

Coverage and what is missing

The IRS bulk XML corpus covers electronically filed returns, mandatory for most filers for tax years beginning after July 1, 2019. Earlier years are real but partial, and small paper filers are absent. Within the posting, the field mapping resolves every publishable field for more than 99% of grant-bearing filings, and parsed 990-PF totals reconcile with the filers' own stated totals within 1% for more than 99.9% of filings that state one. Both figures are published with each dataset version.

Filing lag

A foundation's grants for a calendar year are reported on a return filed the following year and posted by the IRS some months after that. This is a rear-view mirror by construction. The dataset version on every page says which IRS posting it saw.

Donor-advised fund sponsors

The largest grantmakers by dollars are donor-advised fund sponsors — Fidelity Charitable, Schwab Charitable, the National Philanthropic Trust, community foundations. Their grants are the recommendations of thousands of individual account holders, not the priorities of one institution. They are real grants, correctly attributed to the sponsor that paid them, and they should be read with that in mind.

Named individuals

Scholarship, fellowship and hardship payments to natural persons are reported on the same forms. They are tagged at parse time and excluded from the edge list, the pages, and the published dataset's default view. Publishing named individuals would serve nobody.