NumFOCUS revenue was $8,376,226 in 2024, against $11,090,430 of expenses

Quick answer (September 2026). NumFOCUS reported total revenue of $9,289,091 in calendar 2022, $7,245,742 in calendar 2023 and $8,376,226 in calendar 2024. Expenses over the same three years were $5,744,317, $9,951,106 and $11,090,430, so 2022 closed with a $3,544,774 surplus and the two years since closed with deficits of $2,705,364 and $2,714,204. Total net assets fell from $11,501,146 at the end of 2022 to $5,665,364 at the end of 2024, a drop of 50.74%. The share of that money NumFOCUS can actually spend on itself fell much faster: unrestricted net assets went from $9,009,216 to $539,410, down 94.01%.

The 2023 and 2024 returns are not published on NumFOCUS's own public records page. Every figure here comes from the Form 990 XML that NumFOCUS filed with the IRS, downloaded from the IRS bulk archive and audited line by line.

Five different numbers, and where the public answers stop

Search for what NumFOCUS earns and the first page returns five figures: $8.38M, $8,376,226, $7,245,742, $6,785,603 and $2.8M. They are not all wrong, and the differences are not random.

NumFOCUS is the fiscal sponsor behind a large part of the scientific Python stack: NumPy, pandas, Project Jupyter and SciPy among them. It holds and disburses money on their behalf, which is exactly why the headline revenue number is the least interesting figure on its balance sheet.

Start with the source everyone else uses. NumFOCUS's own public records page opens the tax section with a sentence worth quoting in full:

NumFOCUS, on its own public records page: "Because NumFOCUS is a public charity, we are required by the IRS to publish our tax returns for public inspection."

Beneath that sentence sit eleven Form 990 PDFs. The newest is for calendar 2022, uploaded in November 2023.

The 2023 return was filed on November 15, 2024. The 2024 return was filed on October 24, 2025. Neither is on the page. As of September 10, 2026 the page is two filings behind, and the gap is now approaching two years.

That single fact explains four of the five numbers.

Grading the incumbent answers

It is worth being precise about who is right, because most of them are.

ProPublica's Nonprofit Explorer is correct. Its page shows revenue of $8,376,226, expenses of $11,090,430 and net income of minus $2,714,204 for 2024, and all three match the filed return exactly. Cause IQ publishes the same $8,376,226 and is also correct. Growjo's "$2.8M" is an estimate and is wrong by roughly a factor of three.

The interesting case is fossfoundation.info, which publishes $7,245,742. That is not an error either. It is the 2023 figure, and it is stale for a reason worth knowing about.

ProPublica's public API and ProPublica's public web page disagree with each other. The API's list of extracted filings for NumFOCUS stops at 2023. Its list of known-but-unextracted filings is empty. Yet the same API record carries a field naming the 2024 filing's object identifier, and the human-facing page renders the 2024 figures. So anything built on the API, as directory sites generally are, sees a corpus that ends a year earlier than the website does and has nothing telling it otherwise.

The two things nobody publishes are the two filings themselves, and what is inside them.

The audit: 63 checks, 61 reconcile exactly

Before drawing any conclusion from a filing it is worth testing whether the filing adds up. Across the 2023 and 2024 returns, 63 discrete numeric checks were run: Part I revenue and expense components against their stated totals in both the current-year and prior-year columns, Part X assets less liabilities against net assets in both columns, Part XI reconciliations, Schedule A five-year support tables and their footings and percentages, Schedule J officer compensation against Part IX, the eleven Part I line items where the 2024 return restates 2023, four balance-sheet roll-forwards, and twelve Schedule A overlaps where the two filings cover the same calendar year twice.

Sixty-one reconcile to the dollar. That is a good result and it should be said first. In particular, the restatement audit is clean: every one of the eleven Part I figures the 2024 return republishes for 2023 matches the 2023 return exactly, all four balance-sheet roll-forwards tie, and all twelve doubly-published Schedule A years agree. Part IX foots exactly in both years across twenty and seventeen expense lines respectively. The public support test passes comfortably, at 95.76% for 2023 and 94.53% for 2024 against a 33.33% threshold.

Two checks do not reconcile.

Part XI, 2023, off by $3,423. The reconciliation of net assets starts at $11,501,146, applies a deficit of $2,705,364 and a prior period adjustment of minus $421,594, which gives $8,374,188. The return states closing net assets of $8,370,765. There is no other line on Part XI in that filing to absorb the difference. The same check passes exactly in 2022 and in 2024, so this is a single-year footing error, not a convention. On $8.37M it is 0.04%, and it is a rounding-scale problem rather than a material one.

Schedule J, 2024, off by $1. Schedule J reports base compensation of $156,845 and non-taxable benefits of $7,941 for the executive director, totalling $164,786. Part IX line 5 reports $164,785. The equivalent check passes exactly in 2023.

Neither of those is the finding. The finding is a line that is not there at all.

Line 28 is blank

Part X of the Form 990 is the balance sheet. When an organisation follows the standard US non-profit accounting rules, it ticks a box and then completes two lines: line 27, net assets without donor restrictions, and line 28, net assets with donor restrictions.

NumFOCUS ticks the box on both the 2023 and the 2024 return.

On both returns, line 27 is completed. On both returns, line 28 is blank.

For most charities that would be a technicality. For a fiscal sponsor it is the single most important line on the page, because it is the line that separates the money the organisation holds on behalf of NumPy, pandas and Jupyter from the money it can spend on itself. Leaving it blank means the return states a total that its own component lines do not add up to: $539,410 on line 27 against $5,665,364 on line 32 at the end of 2024, and $5,878,835 against $8,370,765 at the end of 2023.

This is not how NumFOCUS used to file. Its self-published 2022 return completes line 28 in both columns: $2,491,930 at the start of 2022 and $2,491,930 at the end. Those columns foot exactly against line 32. So line 28 was filled in for 2022 and left blank for the two years since.

That 2022 return also validates the arithmetic used below. Subtracting line 27 from line 32 on the 2023 return gives a restricted balance of $2,491,930, which matches the printed line 28 in the independently published 2022 document to the dollar. The subtraction works, so it can be trusted for 2024 too.

What the missing line would have said

Derived by subtraction, and stated as a derivation rather than as a printed figure:

Two things stand out.

The first is the collapse in the last column. NumFOCUS ended 2022 holding $9,009,216 it was free to spend and ended 2024 holding $539,410, a fall of 94.01%. Almost all of the two-year deficit landed on the unrestricted side. The restricted balance did not fall at all; it roughly doubled, rising $2,634,024 in 2024 alone.

The second is stranger. The restricted balance is identical to the dollar at three consecutive year ends: $2,491,930 at the end of 2021, 2022 and 2023, two of those printed by NumFOCUS itself. For an organisation whose total assets swung from $8.2M to $12.0M and back to $8.8M across those same years while holding project funds for dozens of sponsored projects, a restricted balance that does not move by a single dollar for three years is difficult to read as a measurement. Either it genuinely did not change, or the figure was carried forward rather than recalculated. Which of those is true cannot be determined from the filings, and it is worth saying why: NumFOCUS has no audited accounts to check it against.

No audit, no review, no audit committee

Both returns answer the governance questions in Part XII the same way. Were the financial statements compiled or reviewed by an independent accountant? No. Were they audited by an independent accountant? No. Does the organisation have an audit committee? No.

That is unusual at this size. It is why there is no Schedule D reconciliation between the return and a set of audited accounts, which for most organisations of this scale is where a second, independently attested revenue figure appears. And it is why the blank line 28 has no external check standing behind it.

One small internal inconsistency follows from the same answers: the 2023 return says the statements were not audited, then ticks a box inside the audited-basis group specifying that they were prepared on a separate basis. Only one of those can be doing useful work.

Two published revenue figures, and a gap that closed

NumFOCUS does publish annual reports, and they carry their own financial section. They do not agree with the tax returns.

For 2022, the annual report states total revenue of $10,366,852. The Form 990 for the same calendar year states $9,289,091. The gap is $1,077,761, or 11.60%. Part of it is visible in the report itself: a fiscal sponsorship line of $698,852 appears identically on both the revenue and the expense side, so the report grosses up an internal transfer that the return does not. That accounts for about 65% of the difference and leaves $378,909 unexplained. The expense totals diverge similarly, $6,785,603 against $5,744,317.

For 2024 the two figures are effectively the same: the annual report states $8,378,968 against the return's $8,376,226, a gap of $2,742, or 0.03%.

That is a real improvement and it deserves to be recorded as one. Whatever changed between the two reporting cycles brought the published narrative and the filed return into line.

The 2024 report is also markedly more detailed elsewhere. Schedule O of the return itemises the largest expense line, other professional fees, and the itemisation covers 71.0% of that line in 2023 against 97.3% in 2024. The unitemised remainder fell from $1,299,263 to $126,634.

Three smaller things in the 2024 report are worth noting for anyone quoting from it. Its revenue subtotals for donations, grants and other income each foot exactly. Its restricted and unrestricted split sums to $8,378,967 against a stated total of $8,378,968, one dollar apart. And its events total is printed as "$2,098,80", which is missing a digit; the two components immediately above it, $1,651,904 and $446,904, sum to $2,098,808.

There is also a presentation oddity outside the numbers. The 2023 annual report is linked from the index page not at a numfocus.org address but at gecko-algae-bt7k.squarespace.com, an unconfigured Squarespace placeholder domain. It resolves and serves the report, but it is not a durable home for an organisation's financial disclosure.

Where the money went

The deficits are not a revenue collapse. Revenue rose 15.60% in 2024. Expenses rose 93.07% across the two years, from $5,744,317 to $11,090,430.

The largest single component is contractor spending on the sponsored projects themselves. Schedule O breaks it out: development and maintenance rose from $2,844,549 to $3,611,728, up 26.97%, with community management appearing as a new $538,707 line in 2024. Salaries and benefits rose from $1,088,774 in 2022 to $2,012,199 in 2024, up 84.8%, against headcount moving from 22 to 25 employees. Reported volunteers rose from 210 to 383.

The expense mix is not the story either. Program services took 94.1% of 2024 expenses, management and general 4.1%, fundraising 1.8%. Whatever else the filings show, the money is going to the work.

The revenue mix did shift. Contributions and grants fell slightly, from $4,640,024 to $4,485,390, while program service revenue rose 46.2%, from $2,492,178 to $3,642,367, driven by event income and service agreements. A $516,324 government grants line appears in 2024 and was absent in 2023.

Liquidity followed the reserves down. Cash and investments went from $8,238,062 at the end of 2022 to $5,406,081 at the end of 2024.

One figure this piece will not give you

The obvious next number is a runway: unrestricted reserves divided by monthly expenses. It is not computed here, and the reason matters.

Total expenses at a fiscal sponsor include spending funded by restricted project money. Dividing unrestricted reserves by a total that includes restricted-funded spending produces a precise-looking figure that is measuring two different pools against each other. Without audited statements separating them, that split cannot be reconstructed from the returns. The honest answer is that the filings do not support the calculation, and naming it and setting it aside is more useful than publishing it.

The same caution applies to the 2022 revenue gap. $378,909 of it remains unexplained, and the documents that would explain it do not exist.

How to reproduce this

Everything above is checkable without an account or a subscription.

NumFOCUS's EIN is 45-4547709. ProPublica's Nonprofit Explorer API returns the organisation record at projects.propublica.org/nonprofits/api/v2/organizations/454547709.json, which is where the 2024 gap in the extracted data is visible.

The filings themselves come from the IRS bulk archive. Each calendar year of received returns has an index at apps.irs.gov/pub/epostcard/990/xml/{year}/index_{year}.csv; grep it for the EIN and the row gives an object identifier and the name of the zip archive holding the XML. NumFOCUS's 2023 return is object 202433209349306003, received in 2024. Its 2024 return is object 202512979349301136, received in 2025. Note that the index year is the year the return was received, not the tax year it covers, so both years have to be searched.

The 2022 return used to validate the subtraction is the PDF NumFOCUS publishes itself, linked above.

If you want the same picture for a comparable organisation, Software Freedom Conservancy's filings tell a similar story about where the public record stops, and NumFOCUS granted Conservancy $40,000 in 2023 for intern sponsorship, so the two appear in each other's paperwork.

What this adds up to

NumFOCUS earned $8,376,226 in 2024 and spent $11,090,430. It has run two consecutive deficits totalling $5,419,568 and has spent down almost all of its free reserves doing it, while the money held for the projects it sponsors grew.

None of that is hidden. It is all on two returns that are filed, public and reproducible from the IRS archive. It is simply not on the page where NumFOCUS says it publishes its returns, not in the API that most directories read, and not derivable at all from the one line the returns leave blank.