Corrections
Corrections
When a figure, a label, or a claim published here turns out to be wrong, the page is corrected and the correction is recorded below with the date it was made. Entries are never edited afterwards and never removed; a correction to a correction is a new entry.
To report an error, write to kian@americanfactbook.org with the page and what you believe is wrong.
The pages of 2,787 organizations that file the short Form 990-EZ understated their largest reported compensation, named the wrong highest-paid person, or named nobody, because the site left out the form's "estimated amount of other compensation".
A Form 990-EZ lists each officer's pay in three columns: compensation (c), benefits and deferred compensation (d), and an estimated amount of other compensation (e). The site added (c) and (d) only, treating (e) like the Form 990-PF's column (e), which is an expense account and is counted in no total. On the 990-EZ it is pay, the counterpart of a Form 990's other compensation, so a filer's largest reported compensation now adds (e) to (c) and (d), and the page names (d) and (e) together beside the total. The 990-PF's column (e) stays out. Of the 722,188 organization pages, 2,787 changed: 1,550 show the same person with a higher total, 158 a different highest-paid person, and 1,079 that showed no paid person now show one. The ranking of organizations by compensation grew from 263,423 to 264,502: the national rank of 234,907 organizations moved, the rank within their state of 163,306 and within their category of 118,466. None of these organizations is in the hub's lists of the largest, so the hub's rankings and totals are unchanged. The figures were reloaded from the stored returns on 14 October, and the counts were measured from the same returns under both rules.
Nine sub-agency pages stated another sub-agency's rank among sub-agencies, and its rank within its department, for fiscal year 2025.
The nine pages were the Departmental Offices of the Treasury and of the Interior, and the Offices of Inspector General of Health and Human Services, Homeland Security, Labor, Justice, the General Services Administration, the Interior and Defense. Each obligations figure was right. The page found its rank by the publisher's code for the office, and every department's Inspector General office shares one code, as do the Departmental Offices. So the rank came from whichever office sharing the code the database returned first. The page now finds its rank by its own address, which is unique. A check compares the rank on every page each spending ranking links to, and each sub-agency's rank within its department, with the ranking's figures. The check the entry below describes compared the ranking's last row only.
The pages of the eight sub-agencies at the foot of this ranking, each with a negative figure for fiscal year 2025, stated a rank two places lower than the ranking did.
The ranking lists only sub-agencies the publisher recorded a non-zero figure against, because a rank by obligations for an entity with no obligations is a claim the data does not make. The sub-agency pages ranked every sub-agency with a page, including two with exactly $0 in fiscal year 2025, the Inter-American Foundation and the International Trade Administration. Both sat above every negative figure, so the eight below them were each two places lower on their own pages, and each page counted two more sub-agencies than the ranking. A page for a $0 year now says no obligations were recorded and states no rank, and the pages rank the same set as the ranking. A check compares every rank the ranking prints with the rank its page states.
The counties ranking by federal awards printed 503 counties, and every rank from 108 down was too high by one or more places. The counties ranking by federal contracts printed 499, counted the same way.
The rankings are counted over the figures committed to this site's repository, which hold the hundred largest counties in the country and the ten largest in each state, not every county. Up to the hundredth row that makes no difference. Past it, a county that is neither is missing from the list, and every county below it moved up a place. Worcester County, Massachusetts, was printed 108th; it is 109th of 3,232, which is what its own page says. The same thing happened to 395 more counties below it. Both rankings now print the hundred largest, whose ranks are exact. A test now reads how many counties those figures hold, from the step that writes them, and fails a ranking that prints more. A check comparing each printed rank with the rank the county's own page states found this.
Four briefs printed thirteen figures that their own queries no longer returned, from 23 September 2026 until 24 September. The largest error named the wrong fiscal year and understated a county's total by $16 billion.
Each figure in a brief is stored twice: the value the query returns, and the text the page prints. On 23 September the briefs' sample data was refreshed to match the hub pages, and the values were recomputed. The printed text was not updated, and no check compared the two. So these pages printed the earlier text beside current data. The county that does not get the most said fiscal year 2026 where the figures are for 2025. It printed Hennepin County's total as $171.9 billion; it is $188.0 billion. It printed 98.7% direct payments as 99.1%, Los Angeles County's $80.1 billion as $68.0 billion, Fairfax County's $41.5 billion in contracts as $32.7 billion, and seven of the top ten counties as eight. The state that gets the most per person printed the Department of Health and Human Services' share of North Dakota's total as 81.5%; it is 91.1%. It printed the department's $89.7 billion as $80.2 billion, the Social Security Administration's $3.4 billion as $3.0 billion, and the District of Columbia's $69,671 per resident as $69,669. Where the money is, on the nonprofits hub, printed the hundred largest organizations' $933.4 billion and 20.7% as $935.1 billion and 20.8%. Most federal spending is not a contract printed grants of $1.24 trillion as $1.25 trillion. Every value and every query was correct; the text beside them was not. The text is now corrected, and a check that every printed figure matches its value is being added to the build.
On 42 crime and state pages, a figure whose next digit was a 5 could print its last digit one unit off, from the first deploy on 19 September 2026 until 24 September. The committed figures were right throughout.
The site's build tool copied our committed figures into the deployed site by reading each file and writing it out again, and in doing so moved 4,137 of them by the smallest amount a computer's arithmetic can represent — one part in about ten thousand million million. Almost always that is invisible. Where a figure sat exactly on a rounding boundary it was not: Florida's 2025 burglary rate is 120.74999999999999 per 100,000 in the committed figures, which rounds to 120.7, and the deployed copy held 120.75, which the page printed as 120.8. Forty-two pages were affected, all showing crime rates: the burglary, homicide, motor vehicle theft, rape and robbery offense pages and their state rankings, seventeen crime state pages and fifteen state profile pages. The committed figures, the queries behind them and every check made against them were correct the whole time; what was wrong was the copy the site served. The site now reads the committed files byte for byte instead of copying them, and a check on every build compares every number the build contains with the committed file it came from, so a copying step cannot change a figure again without failing the build.
Ten sub-agency rows shared three addresses, so seven sub-agency pages showed another department's figures under a name five departments have.
A sub-agency's page address was built from its name alone. Five departments each have an Office of the Inspector General, three have one spelled without the "the", and two have Departmental Offices, so ten rows resolved to three addresses: 216 sub-agencies and 209 pages. The page then at that name alone showed the Department of Justice's $5,686,751.67 while the ranking linked five rows to it, including the Department of Health and Human Services' Office of the Inspector General at $333,045,883.43. A reader following the row for one department read another department's number under the right name. The address now carries the parent agency where the name is not already the agency's own, so that page's address now names the Department of Justice and each of the ten has its own. The pages were linked for about an hour, from the entity links going live at 21:45 UTC to the fix; no sitemap ever advertised them. The same change fixed a second symptom of the same cause: a page said "ranks 107 of 218" while its index listed 209, because one figure counted rows and the other counted pages. Both now count the set the index lists, the page says what its denominator is, and a check run after every deploy asserts the two agree so they cannot drift apart again.
Historical crime rates weighted a year's twelve months equally when the share of each state that reported changed within the year, overstating the early 1990s peaks.
The 18 September correction fixed the denominator: a rate divides by the population the reporting agencies served, not by the whole jurisdiction. That was right, and how the twelve months were combined into a year was not. We divided a year's offenses by the population times the average of its twelve monthly coverage percentages. An average is the correct weight only when the months are alike, and in the years when reporting was patchy they were not: Alabama's 1993 coverage ran from 49.0% to 95.8% and its monthly offense counts from 554 to 19,880, concentrated in the months when most of the state reported. Alabama's 1993 violent crime rate was shown as 1,411.2 per 100,000 against the FBI's own published 1,065.2, and the state page described the fall since as 75.0% where the publisher's figures give 66.9%. The site now publishes the FBI's own annual rate — the sum of the twelve monthly rates it published — and publishes no rate at all for a year it did not report in full. 925 of 21,320 state, offense and year combinations moved by more than 1% and 100 by more than 25%, all of them before 2012; no figure for 2024 or 2025 changed, which is why the error was invisible on the pages that are read most. The check that would have found it already existed and reported 478 combinations more than 1% apart, but it was rendered on the methodology section as a figure about our arithmetic rather than asserted as a gate. It is now a gate, and it compares the publisher's counts, coverage and population against the publisher's own rate rather than comparing our arithmetic with itself.
Crime rates were divided by the population of the whole jurisdiction rather than by the population the reporting agencies served, which understated every rate.
Until this correction, every crime rate on this site divided the FBI's offense count by the whole population of the jurisdiction. The FBI's counts are only what reporting agencies submitted, and the FBI publishes the population those agencies served in the same response — we were using the jurisdiction's whole population instead. Each rate was therefore understated by exactly the share of the population that did not report: slightly in most states, by a third or more in the least-covered. Florida's 2021 violent crime rate was shown as 95.7 per 100,000 residents where the FBI's own figure is 203.8. Rates now divide by the population the reporting agencies served, which is the denominator the FBI itself uses, and that figure is published beside every rate. The correction moved 35 of 51 jurisdictions in the 2025 violent crime ranking, and the states that moved furthest are those with the thinnest reporting, which the previous arithmetic pushed towards the bottom of the table. The national violent crime rate for 2025 moves from 317.6 to 329.3 per 100,000 residents, and the property crime rate from 1,496.8 to 1,552.1. The pages affected are the crime hub, the three state rankings and all 51 state pages. The test that should have caught this asserted the same formula the code used, so it compared the implementation against itself and passed; it has been replaced by one that compares our arithmetic against the rate series the FBI publishes, across every state, offense and fully reported year.
Updated