CharitySense Data

How we rate nonprofits, and what the ratings can tell us

We went back to the current scoring rules and the wider evidence now available for each organization. The most useful lesson is to read the score beside its filing, its coverage, and the facts the score does not try to compress.

By the CharitySense Data team

18
possible measures for a full Form 990
5
weighted categories
3
filing types with a scoring policy
0–100
the score shown on a profile

01/score begins with

A score begins with a narrower question

When we open a nonprofit profile, we want to know what the available evidence supports. A score can summarize parts of that evidence, but it cannot tell us whether a program worked, whether a restricted gift fits a donor’s purpose, or whether every named partner is dependable. We now have more sources to explore those questions, including public reporting, relationships, leadership records, and federal awards. Their presence makes the limits of a single number easier to see.

Our rating uses a set of published choices about stewardship and disclosure. The score is calculated when a profile is opened from the organization’s current filing and supporting evidence. We do not treat an older batch of scores as a description of what the current policy would produce. That matters here: an earlier version of this article displayed a historical score distribution and called it the live rated population. We removed that chart because it no longer answered the question it appeared to answer.

A useful score points back to evidence. The most revealing question is often which evidence is available to score.

02/choices inside a full-return

The choices inside a full-return rating

A full Form 990 can feed 18 possible rating measures across five categories. Financial efficiency carries 40 percent of the final score, transparency and controls 30 percent, accountability 20 percent, and sustainability and growth 5 percent each. The chart opens into the individual measures. Some are relevant only when the organization reports foreign activity, grants, or related entities.

Chart source values; units and population are described in the surrounding figure.
nameweight
Financial efficiency40
Transparency and controls30
Accountability20
Sustainability5
Growth5

Fixed share of the full Form 990 score, in percent. Select a category for its possible measures and their relative weights. Source: CharitySense evidence-policy candidate, October 2026.

The numbers inside a category are relative weights. Mission deployment has a weight of 25 next to program expenses at 15 and fundraising efficiency at 5. It does not automatically contribute 25 points to the final 100. The applicable measures are combined first, then that category receives its 40-percent share. This distinction is especially important when a measure does not apply or lacks a usable value.

Official-site evidence about monitoring, governance, policies, financial reports, outcomes, and named relationships is now a separate public-disclosure measure, outside the overall rating. A published claim counts as evidence that an organization made the claim. It does not independently prove the control worked or the claimed outcome occurred.

03/missing measure

A missing measure is not always a zero

The rules distinguish a reported zero, a required item left blank, a question that does not apply, and a fact we simply have not observed. Those cases should not lead to the same conclusion. Applicable but unknown measures retain their category denominator weight and earn no evidence-supported points; they are not called explicit failures. A whole category with no usable score retains its share of the overall denominator. That prevents a narrow slice of disclosure from looking like a complete picture.

For Forms 990 and 990-EZ, a publishable score also needs a supported mission-deployment calculation. Across the scored filing types, at least half of the category weight and 35 percent of the eligible input opportunities must be covered. When the evidence falls short, the profile says it lacks enough data rather than issuing a confident grade. Confidence can also be reduced or blocked by problems with the underlying facts. We read that coverage and confidence alongside the number.

This is a substantial distinction in the present corpus. We have about 950,000 organization records prepared with rating reference inputs, alongside more than 2.2 million public profiles. A prepared input record is not a published score: the form-specific evidence and coverage rules still have to be evaluated when someone opens the page. Counting prepared records as rated charities would overstate what the data can establish.

A blank website signal deserves care
Our current research includes about 33,500 organization website records, while the public profile count is 2,222,250. These counts come from different sources and do not measure an overlap rate. The disparity is a reminder that a missing public-site signal often means we have not observed the relevant page. It should not be read as proof that a policy or practice is absent.

04/filing form changes

The filing form changes the question

A direct-service charity and a private foundation do different work. Scoring both by a program-expense ratio would blur that difference. Our full Form 990 rating has 18 possible measures; the shorter Form 990-EZ has 11 across four filing-supported categories. A foundation’s Form 990-PF uses six possible measures under its own five-part model, emphasizing grant deployment, cost efficiency, transparency, compliance, and endowment health. These are policy inventories, not counts of measures present for every organization.

ReturnHow we read it
Form 990The full stewardship model, with 18 possible rating measures.
Form 990-EZAn 11-measure, four-category version using the shorter return’s own evidence.
Form 990-PFA separate six-measure foundation model centered on grants and endowment health.
Form 990-NAn e-Postcard; too little financial detail for this rating.
Form 990-TA return about unrelated business income, not an organization-wide rating.

Source: current CharitySense rating policy and IRS descriptions of the returns. A possible measure may be inapplicable or unavailable for an individual filer.

The distinction is especially useful for small organizations. Form 990-N is a brief annual notice, so an unrated e-Postcard is not a failing grade. Likewise, a Form 990-T may appear alongside a full return, but its unrelated-business figures do not stand in for the organization’s whole operation. The form, year, and evidence width belong in the conversation before the score does.

The current filing inventory includes 1,540,498 Form 990-N submissions, about one in six of the 8,868,546 filings in the corpus. Those are filings across years, not distinct organizations. Still, they show why the difference between unrated and poorly rated matters to a large number of readers.

05/What the broader record

What the broader record adds beyond the score

The dataset now holds roughly 456,000 organization-level relationship records and 26,900 organization records with federal-award information, in addition to filing histories and public-site research. Those are record counts, not counts of every partner or award in the nonprofit sector. They let us ask more specific questions: who appears in a filed grant or structural relationship, what federal activity is associated with an organization, and whether a public description lines up with those records. A difference is a reason to look closer, not a finding of wrongdoing.

These newer sources do not all become rating inputs. We think that boundary is useful. Grant eligibility checks, restrictions, screening results, leadership connections, and federal awards answer questions that a stewardship index cannot settle. For a grant decision, our grantee due diligence guide follows those separate checks. To see why a low overhead number does not tell the whole story, read the overhead question.

More evidence should make the reading more precise. It should not turn every available fact into another point on a scoreboard.

06/How we read

How we read a profile now

We begin with the return type and tax year. Then we look at whether the rating is supported, which categories have evidence, and where the largest category weights sit. A low program-spending measure might reflect an accumulation year, a foundation’s planned grant cycle, or an unresolved filing fact. The underlying revenue, expenses, grant detail, and validation checks help distinguish those cases. We use public reports and relationship evidence for context rather than assuming the score explains them.

There is another subtle point: the overall 0-to-100 number is not a percentile rank, even though some open-ended ingredients use peer percentiles to get onto a comparable scale. A score of 80 does not mean the organization outranks 80 percent of peers. It means the weighted, applicable evidence produced 80 points under this published policy. Our nonprofit-vetting walkthrough shows how to move from that summary into a closer reading.

Behind the Data

We checked the evidence-policy candidate in October 2026 and retained the source-inventory observations from September 25, 2026. The 18, 11, and six measures are potential overall-rating inputs for the three rated filing types; website publication is separate. They are not a claim that every organization has every measure. Category weights and minimum coverage rules come from the methodology. The public-profile and Form 990-N filing totals come from a site-wide snapshot computed September 17. Rating-reference, website, relationship, and award record counts are approximate inventory counts. They describe the reach of different sources, not complete coverage of the sector or a count of scored organizations.

We deliberately do not reuse the historical rating histogram or the old count of rated organizations. The current score is calculated when a profile is requested, so a past run cannot establish today’s score distribution. IRS filing years arrive with a delay, churches and some related organizations are generally excepted from annual filing, and the newest records may still change. For the full metric definitions and form-specific rules, see our methodology.

Sources

Ask the research team

What would you want us to pull from the filings? Tell us, and a future piece may answer it.

07/Final Thoughts

Final Thoughts

We want a rating to be a starting point for inquiry. The current model makes its priorities visible, separates filing types, and withholds a firm score when too much evidence is missing. The expanded public and relational record gives us better ways to investigate what sits behind a number. It also reminds us that some important questions should stay outside the rating entirely.

About the author

The CharitySense Data team

Research team

We read public nonprofit filings and other sources to understand how the sector works. This blog shares what we find, along with the limits of the evidence. Each article explains how its figures were measured and cites the records or primary sources behind them.