What 800 nonprofit filings show about proving your own numbers

I spent a week reading the public filings of 800 American nonprofits, and I want to describe one pattern I did not expect to find, along with the thing that made it visible.

The organizations are all mid-sized — between five and twenty-five million dollars of annual revenue — working in human services, education, youth development and international development, spread across 54 states and territories. I pulled every Form 990 available for each, up to thirteen years apiece.

I was looking for something narrow: whether the numbers these organizations report to their funders can be traced back to the systems that produced them. What I found first was something simpler, and stranger.

Revenue that moves without the business moving

A Form 990 splits incoming money into a few lines. Two of them carry almost everything a service organization receives. Contributions and grants covers donations, foundation grants and government grants. Program service revenue covers money earned by delivering the mission — fees, contracts, payments for services rendered.

In 118 of these organizations — 16% of those with at least five years of comparable filings — one of those two lines moves by at least 25 percentage points of total revenue in a single year, while total revenue stays essentially flat.

Not the organization growing. Not a funder arriving or leaving. The same money, landing on a different line.

One organization reported 93 to 96% of its revenue as contributions for eleven consecutive years, then 5% the following year — with total revenue slightly up. Another sat at 98–99% for nine years, then dropped to 11%, while revenue rose by five and a half million dollars. A third has crossed that boundary in both directions, twice, and currently sits roughly where it started a decade ago.

Thirteen organizations oscillate — the line moves one way, then back again.

Why this is not an accounting scandal

I want to be careful here, because the obvious reading is wrong.

There is a genuine, well-known ambiguity underneath this. When a government pays a nonprofit, someone has to decide whether the government is buying a service — an exchange transaction, which belongs in program service revenue — or funding a mission without receiving commensurate value in return, which makes it a contribution. For Medicaid-funded services, state contracts and pass-through county money, that question is arguable in good faith.

It was arguable enough that in 2018 the Financial Accounting Standards Board issued a standard, ASU 2018-08, specifically to reduce the inconsistency. It pushes in one direction: most government grants and contracts are contributions, because the government does not receive commensurate value.

So a reasonable person would expect these moves to cluster around 2019 to 2021, when that standard took effect, and to run in one direction.

They do not.

More than half the moves predate the standard entirely. They run in both directions. And when I read the audited financial statements behind a sample of these organizations, only two of twenty-three reports mention ASU 2018-08 at all — one of them as boilerplate about a pronouncement not yet adopted.

I also went looking for the most obvious false positive. Federal relief money arriving in 2020 and leaving again produces exactly this signature: a line jumps, then returns. I found 30 organizations in that pattern and excluded every one of them. The 118 are what remains after that.

So this is not fraud, and it is mostly not a standard being adopted. It is something less dramatic and more interesting: a judgment call that gets made again each year, by whoever is closest to the filing at the time.

Why it matters more than it looks

If this were only about presentation, it would be a curiosity. It isn’t, for three reasons.

The public support test. A public charity has to demonstrate, on a rolling five-year lookback, that its money comes from a broad base. Contributions and program service revenue are weighted differently in that calculation. A line that moves eight million dollars inside the lookback window doesn’t just make year-over-year comparison awkward — it moves an input to the test that determines the organization’s tax status.

Funder reporting. Grant reports, board packets and renewal applications are built on multi-year comparisons drawn from these same categories. If the underlying definition changed in year three and the comparison wasn’t rebuilt, the report quietly compares two different things. Nobody notices, because the report is assembled by a person who knows what they meant.

That last clause is the actual finding. A number that shifts category without a documented reason is a number that no system is producing. It is being decided. And a decision that lives in one person’s judgment rather than in a rule is a decision that leaves when they do.

The document nobody reads

Partway through this I realized I was using the wrong source.

The Form 990 is the famous one — public, searchable, the basis of every charity rating site. But it is a derived document, and it is old. Because organizations take the automatic six-month filing extension and the IRS then needs months to process and publish, the most recent fiscal year with near-complete coverage in my sample was 2023. Only 12% had a 2024 filing available.

Meanwhile, any organization spending a million dollars or more in federal awards must file a Single Audit — a full financial statement audit plus a compliance audit against each major federal program — with the Federal Audit Clearinghouse. Those are due nine months after fiscal year end, with no extension, and they post quickly.

In a 250-organization sample, 42% had a Single Audit on file. Of those, 82% had a 2024 or 2025 audit available — against 12% with a 2024 Form 990.

The audit package contains the audited financial statements the 990 is derived from, the notes that explain the accounting policies, every finding the auditor raised, and the name and title of the person who signed for the organization. It is free, public, and available through an API.

It is a strictly better document, roughly two years fresher, and almost nobody outside the compliance world reads it.

The most interesting thing I found came out of that comparison. One organization’s audited statements state plainly that it adopted ASU 2018-08, under which grants and contracts are treated as contributions. Its Form 990 for the same fiscal year shows contributions falling from 100% of revenue to 20%. Two documents, one year, one organization, pointing in opposite directions.

I am not naming it, and I want to be explicit about why. There is almost certainly an explanation, the organization has done nothing wrong, and publishing a list of charities whose filings look odd would be a way of manufacturing alarm in order to sell a remedy. The pattern is worth writing about. The names are not mine to publish.

What I think this says

Every organization here has an auditor, a finance lead and a board. The inconsistency isn’t a competence problem, and it isn’t a software problem — most of them have perfectly good accounting systems.

It is a seam problem. Financial data lives in the accounting system. Delivery data — who was served, how often, with what result — lives in case management systems, service logs, timesheets and spreadsheets. Funder reports need both, cut on calendars that match neither the fiscal year nor each other. Federal awards run October to September, many state contracts July to June, foundations on their own grant anniversaries.

Nobody’s job title covers the join. So it gets done by hand, by someone who is good at it, every reporting cycle, forever. It works. The reports go out. The audits come back clean.

And the cost of it is invisible, because it’s absorbed rather than spent — until the person who knew how the spreadsheet worked takes another job, and the organization discovers that a decade of its reported history rested on their memory.

The analysis used the ProPublica Nonprofit Explorer API for Form 990 data and the Federal Audit Clearinghouse API for Single Audit data. Both are public and free. Figures are counts from the sample described, not national estimates.

I do independent assessments of exactly this problem for nonprofits — whether the data can prove what the programs actually do. If it’s relevant to you, my email is ravi@helloravi.com.

Ravikanth Andhavarapu assesses whether an organization's data can prove what it actually did. What that involves · ravi@helloravi.com