This foundation directory curates metadata that is rigorously organized enough for academic research, and is easy enough to maintain by open source organizers. We encourage contributions!

What We Collect

We record non-profit FOSS organizations, the people who lead them, and how they are funded. Data about code, packages, repositories, and similar technical details is left to ecosyste.ms and similar services, which we link to rather than copy. Every fact should trace to a source, preferably the organization’s own website. How carefully each dataset is checked, and what we record about its sources, depends on what the dataset is for; see How We Source Data and the rules for data.

Data Organization Philosophy

All data, basic instructions and explanations about the site, and basic visualizations or listings should be stored in human-editable formats, and should be served statically with minimal dependencies. This ensures that maintenance issues don’t break things, and that people with limited browsing clients can still participate. Advanced features or larger data analysis tools should seek to minimize dependencies or any regular required maintenance. Our data should also strive to be consistent and persistent. This helps both casual users find the data, and helps researchers by keeping stable URLs and formats.

Datasets

Each dataset has its own page, which documents its fields, schema, and tools. Hand-edited datasets are maintained by people through pull requests; generated ones are produced by our tools from other data and are never edited by hand. This list is kept in _data/datasets.yml, and is also published as a DCAT catalog at /api/v2/catalog.jsonld for data portals and other catalog tools.

Dataset What it holds Where it lives Schema Maintained
Foundations One record per non-profit FOSS organization: legal identity, governance, policies, funding _foundations/ foundations-schema.json hand-edited
Entities Self-organized projects that are not legal corporations, and typically use a fiscal host, such as CNCF _entities/ work in progress hand-edited
Sponsors Key sponsors that fund several foundations: the domains that join them to sponsor data, and lookup identifiers such as LEI, SEC CIK, and Wikidata _sponsors/ sponsors-schema.json hand-edited
Sponsorships Ordinal levels of recurring sponsorship programs, drawn from each organization’s own prospectus _sponsorships/ sponsorships-schema.json hand-edited
Sponsor lists and history Current sponsor lists, monthly history with provenance, and summary reports _data/sponsorships/, history/sponsorships/ described on Sponsorship Models generated
Platform crosswalk Which code host and package registry accounts belong to a foundation, with evidence _data/crosswalk/ crosswalk-schema.json hand-edited; its JSON-LD exports in api/v2/crosswalk/ are generated
Hosted projects Projects hosted by some foundations, for context _data/projects-<identifier>.csv CSV, with column names in the header row scraped, then reviewed by hand
Crosswalk evidence and attestations Why each crosswalk link is believed: the evidence items inside each crosswalk file, the attestors whose confirmations count, and records of the people who confirmed links evidence in _data/crosswalk/; _data/attestors.json; public person records in _data/attestations.json, private ones in the maintainers’ private repository crosswalk-schema.json, attestors-schema.json, attestations-schema.json hand-edited; attestors by maintainers only; evidence with fdncrosswalk attest, person records with fdncrosswalk add-person
ProPublica 990 tax data ProPublica Nonprofit Explorer’s summary of IRS 990 filings for US foundations: about 70 totals per filing, a year or more after filing _data/p990/ ProPublica’s format generated (downloaded by foundation_reporter)
IRS 990 tax data Line-item detail parsed from the IRS’s own 990 e-file XML, including recent years ProPublica has not parsed yet not yet published here Frictionless Data package written by the tool in progress: generated by foss990parser
Budgets Planned: actual or estimated budget figures from organizations’ own reports not yet not yet planned

Entities are self-organized and funded projects that are not legal corporations. Key examples are CNCF (a division of the Linux Foundation) and Debian (with a documented governance and budget structure, but uses SPI as a fiscal host).

US 990 tax data comes from the IRS Form 990 that US nonprofits file each year, recorded from two different sources, so it is kept as two datasets. ProPublica 990 data quickly gives a solid financial overview of many US-based foundations; we host ProPublica’s organizational 990 JSON as a convenience, and expect to keep its URLs permanently. IRS 990 data is parsed directly from the IRS’s e-file XML, with much more detail and newer years, but in a different format with some subtly different values. The US 990 Taxes Model explains both and how they differ; most researchers need only one of them.

Crosswalk evidence records why we believe each crosswalk link. Most evidence is a public page anyone can check again; human-attested evidence is a confirmation that a trusted attestor got out of band from someone at the organization. These attestations are given by volunteers and have not been legally reviewed to qualify as attestations under the EU Cyber Resilience Act or other legislation; see their legal status. Records of those people are kept private by default, and are public only when the person agreed.

Non-US Tax Data Some countries post nonprofit financial data; most do not. Our foundation_reporter tool can download UK Charity Commission data, stored as _data/taxes/uk-<charity number>.json, and we are looking for help in modeling foundations outside the US! This UK data is not yet listed as a dataset: once we have more examples and a parser for it, we will add it as a proper dataset.

How We Source Data

No dataset here is filled in by guessing, but our datasets are not all used the same way, so they are not all checked the same way. A slightly wrong board size on a foundation page misleads a reader a little. A wrong claim that a foundation controls a package namespace could mislead someone deciding which releases to trust. So each dataset gets the level of checking its use calls for, and stores only the sourcing that level needs. CONTRIBUTING.md gives contributors the same rules, and AI agents follow them in AGENTS.md.

Dataset Accuracy it needs, and why How changes are checked What sourcing we store, and where
Foundations, entities, hosted projects Good enough for browsing and general research. Most fields are plain facts, such as a website, a founding year, or a board size, that a reviewer can check at a glance; hosted project lists are best-effort snapshots. A maintainer checks each change against the organization’s own site, which the pull request links. Descriptive text also follows the officers get a say rule. No per-field sources or check dates. The pull request, with its links, and git history are the record: git log _foundations/<identifier>.md lists every change to one record.
US 990 and other tax data Exactly as published. Figures come from government filings or a trusted aggregator such as ProPublica, and we copy them, never correct them. Our tools download the data; review checks that the copy ran cleanly, not each figure. When and how we copied it: the tool that downloaded it, the source named inside each file, and the git commit that added it. See how we source tax data.
Sponsorship models, sponsor lists and history, sponsors Approximate, for rough comparison. Foundations publish very different amounts of detail, so our models can only approximate each program; each page says how approximate. Models are written by hand from each organization’s own sponsorship pages. Sponsor lists are scanned by our tools. Sponsor records are looked up in authoritative registries such as GLEIF and SEC EDGAR. The dates and method of each scan: when each sponsor list was first seen and last confirmed, and where each past list came from, such as an Internet Archive capture. Models link the pages they model and the date their levels took effect. Sponsor records list a source and date for each lookup value.
Platform crosswalk The highest. Some links are read as a foundation’s official control of an account or package namespace, so users expect them to be right and to have been checked by people. Every link needs evidence that anyone can check again, its confidence is computed from that evidence, only listed attestors can confirm a link out of band, and a maintainer reviews every change. Evidence items on every link: its type, a source URL, the date observed, and who recorded it, plus records of attestors and of the people who confirmed links. Once a reviewer has checked a link, it also records who and when; a link without that has not yet been checked. See crosswalk accuracy.

Data not yet published, such as budgets and IRS 990 data, will follow the row closest to it: budgets from organizations’ own reports are approximate, and IRS 990 data is copied exactly.

Licensing

Our licensing philosophy is to allow maximal reuse by anyone, both for datasets hosted, website presentation, and any automated tools.

Code and prose are licensed under Apache-2.0. New datasets and data exports, such as the sponsors records, the public attestation records, and the crosswalk exports, are CC0-1.0. Existing data files are currently Apache-2.0. Each file’s license is declared following the REUSE specification in REUSE.toml.

Citing This Data

Please cite the page or file you used, the date you accessed it, and, for exact reproducibility, the git commit of this repository. We keep URLs and identifiers stable so that citations keep working. We plan to publish citation metadata for each dataset, so that it can be cited directly.

Schemas And Validation

Each dataset’s schema is listed in the table above, and field descriptions live in each schema. Foundation records are also described by the v1 OpenAPI definition. The crosswalk, including its evidence, is checked by its own validator, fdncrosswalk validate, and sponsor records by sponsors.rb --check. Published API files are described on OpenAPI and API versions. Suggestions for improvements to make the data more useful are appreciated! Plans for further linting and validation are on Workflows / Help Wanted.