Corrections
Corrections
We correct errors openly. A correction fixes a fact that was wrong. A clarification fixes wording that could mislead. An update records that the world changed after we published. Each is noted on the post with its date. We never quietly rewrite a claim. While we check a reported error, the post carries a dated note at the top saying it is under review; that note stays, and the correction or update that settles it follows it.
Report an error
The report form is not open yet. Every correction and clarification we make is noted, with its date, on the post it changes. The newest are listed below, and all of them are listed by month.
Corrections log
The 20 most recent of 62, newest first. Every one is listed by month below.
Corrected
Voting and the Democratic Process: An earlier version of this post said 84.3% of eligible voters were registered in 2012. That figure is 2012 turnout among registered voters. Census Bureau survey data show about 71% of voting-age citizens were registered. The post also said about 57 percent of eligible voters cast ballots in the 2016 presidential primary. It now gives turnout in the 2016 general election: about 60 percent of eligible voters, per the UF Election Lab.
Corrected
Corrections Reform: What Works, What Fails, and What It Costs: An earlier version of this post credited the Bureau of Justice Statistics for a count of roughly 2 million people behind bars at the end of 2023. The bureau counts about 1.85 million people in state or federal prisons or local jails at that date. Nearly 2 million is the Prison Policy Initiative's 2025 count, which adds immigration detention, juvenile facilities and other confinement. The post now gives both figures, each credited to its source.
Corrected
Drug Testing Welfare Recipients Is Illogical: An earlier version of this post gave Oklahoma's rate of positive drug tests as 8.8 percent. The Snopes article it cites gives 8.9 percent: 297 of the 3,342 applicants the state screened from 2012 to 2014 tested positive. The post now says 8.9 percent.
Corrected
Corrections Reform: What Works, What Fails, and What It Costs: An earlier version of this post showed an image that labeled the Prison Policy Initiative's estimate of at least $182 billion a year "Total corrections costs" and carried a drafting subtitle. The image and its note are now replaced by figures labeled as their sources define them and a chart of reincarceration rates.
Corrected
The Party Realignment: How Democrats and Republicans Changed Over Time: An earlier version of this post said the South moved from presidential voting to congressional politics, "then state legislatures and local offices," in the Stage 5 text and the key takeaways. We found no source for local offices, and both now end at state legislatures, which a study of Southern elections (McKee) shows following Congress. The 1968 caption and timeline entry said Wallace carried "five Deep South states"; the five were Alabama, Arkansas, Georgia, Louisiana and Mississippi, and both now say five Southern states and name them. The 1932 caption and timeline entry said the South voted Democratic "alongside northern urban centers," which no source cited here supports; both now give the National Archives' state results: Roosevelt carried every former Confederate state and every state but six. The 1964 caption and timeline entry said the Deep South broke toward Goldwater while the rest of the country voted heavily Democratic; Goldwater also carried Arizona, his home state, and both now say so. The update of the same date that cites a source for the 1964 result calls its wording unchanged; the Arizona change was made after that update.
Corrected
The State of the 2026 Economy: Why It Feels Worse Than the Data: An earlier version of this post said "paycheck-to-paycheck can reach well into six-figure households in high-cost regions." We found no source for "well into", and the Federal Reserve survey now cited does not break its figures out by region. The post now gives what the survey found: in 2024, 22% of adults with family income of $100,000 or more spent about as much as their income in the prior month, and 12% spent more. It also said many insured families "delay care, skip specialist visits, or take on medical debt." The survey measures care people went without, not care they delayed, so the post now says they skip care, including visits to a doctor or specialist, or take on medical debt, and gives the same survey's figures: 26% of insured adults went without some medical care in 2024 because they could not afford it, and 17% of all adults had medical debt. And it said that "a large share of imports are inputs into domestic production" and that sweeping tariffs "can raise costs for U.S. firms as much as they raise sticker prices for households." The Congressional Budget Office's outlook, which the post cites, estimates that foreign exporters absorb 5 percent of tariff costs and that the 95 percent borne in the United States reaches consumers: in the near term, businesses absorb 30 percent of their import price increases and pass 70 percent on, and competing domestic producers raise their prices enough that, on net, consumer prices rise by the full 95 percent. The post now says that many imports are used as inputs in domestic production, so sweeping tariffs raise costs for U.S. firms as well as sticker prices for households, and cites CBO.
Clarified
Should Religious Institutions Pay Taxes? A $100B Transparency and Opportunity Cost Case: Three statements are now more precise. The post said many churches are not required to file the same annual public financial disclosures as comparable nonprofits; under IRS rules no church has to file the annual information return (Form 990 or 990-EZ) that other tax-exempt organizations must generally file once their receipts normally exceed $50,000, and the post now says so. The working estimate of $100 billion was labeled as 2026 dollars in three places: the key takeaways, the working-estimate line and the stat row. By CPI-U it is the 2012 estimate in 2025 dollars, about $103 billion at August 2026 prices, and all three now say so. The SNAP line assumed $2,500 per recipient-year; benefits averaged $187.94 per participant per month in fiscal year 2025 (USDA Economic Research Service), about $2,255 a year, so the assumption is on the high side and its count of about 40 million recipient-years on the low side (about 44 million at the actual average). That assumption and its count are unchanged.
Corrected
Should Religious Institutions Pay Taxes? A $100B Transparency and Opportunity Cost Case: Three statements were wrong or had no source. The source line under the 2024 lobbying figures repeated the counts the post first gave, 18 clients and 58 lobbyists, which cannot be confirmed; it now gives the counts OpenSecrets shows, 21 and 62. The housing voucher line assumed an "all-in" annual public cost of $12,000 per household. HUD's 2024 budget request put voucher renewals at $27.8 billion for about 2.3 million households, plus $3.2 billion in administrative fees, so the all-in cost was about $13,500 per household; $12,000 is close to the renewal cost alone (about $12,100). The line no longer says "all-in" and gives HUD's figures. The health center section called $1,200 per patient-year "a conservative benchmark." No source supports that: KFF reports $49.6 billion in health center revenue for 32.4 million patients in 2024, about $1,530 per patient, so $1,200 is below what health centers spend and its count of about 83 million patient-years is high. The post now calls $1,200 an assumption and gives the KFF comparison, about 65 million patient-years at $1,530.
Clarified
Is Universal Basic Income Inevitable?: This post said federal low-income programs differ by "eligibility, application processes, recertification, and enforcement," and that a raise can leave a household worse off "when a household loses multiple benefits at once." Its sources support narrower wording. The post now says programs differ by eligibility, application processes and the conditions for staying eligible, as the Government Accountability Office reports, and that a raise leaves a household no better off, or worse off, when the benefits it loses equal or exceed the raise, which is how the Department of Health and Human Services defines a benefit cliff.
Corrected
The Party Realignment: How Democrats and Republicans Changed Over Time: Three of the four chart images did not match their sources, and the fourth depended on one of them. The two Black party identification charts began in 1948; the American National Election Studies (ANES) series, which they now use, begins in 1952. The second chart's caption credited a post-1964 rise in strong Democratic identification; that share jumped in 1964 and 1968, fell back, and rose again from 2008. The South's Republican presidential vote chart was off by up to 3.5 points; it now uses MIT Election Data and Science Lab (MEDSL) returns: 49.7% in 1992, including D.C. The gap chart now uses MEDSL's House returns. The text said the presidential-minus-House gap held near 10 points through the 1970s and 1980s and closed only after 1994; it was 8.6 to 19.7 points from 1976 to 1988, 2.9 in 1992, and between -4.3 and 2.4 from 1996 to 2016. A 1980 caption calling the South a reliable Republican bloc was also corrected.
Corrected
Why Local Journalism Is a National Security Issue: An earlier version of this post said Russian operatives in 2016 created fake local news sites promoting divisive content. The Senate Intelligence Committee found that the Internet Research Agency ran Twitter accounts designed to look like local news aggregators, such as @OnlineMemphis and @TodayPittsburgh, which posted links to legitimate news sources and tweeted about local issues. The post now says that, and the table of threats says accounts, not sites.
Corrected
Why Local Journalism Is a National Security Issue: An earlier version of this post said studies find "higher rates of misappropriated public funds," "increased cost overruns on public projects" and "higher rates of candidates with previously undisclosed scandals," and that incumbents are reelected more often "even when incumbents are ineffective." We found no study behind these four statements; the study cited on incumbents finds more reelections but does not assess how incumbents performed. The first two are replaced by what the Gao, Lee and Murphy paper found: higher government wages and deficits, and more costly borrowing choices. The other two are cut. The opening said that within months of the Vindicator's closing, city spending irregularities and school board controversies went unreported and local elections featured candidates nobody had scrutinized. No source supports this, and it is cut; the opening now cites a 2020 University of North Carolina report that Youngstown was the first city of any size in the country to lose its sole surviving daily newspaper. The post said that during Hurricane Maria local reporters tracked power, water and passable roads, national media left after a week, and local journalists stayed, documenting recovery failures that became a national scandal. It now says what news reports support: the only radio station on the air through the storm relayed messages between families, and a lawsuit by CNN and Puerto Rico's Center for Investigative Journalism led to the release of partial death records. The post also said national media "parachutes in for a day before leaving"; it now says local reporters are still there after national crews move on.
Clarified
The State of the 2026 Economy: Why It Feels Worse Than the Data: When this post was published on March 15, 2026, it gave the 13% average tariff rate for the end of 2025 without noting that on Feb. 20, 2026, the Supreme Court had ruled in Learning Resources, Inc. v. Trump that the International Emergency Economic Powers Act (IEEPA) does not authorize the President to impose tariffs, or that the IEEPA tariffs had then been removed. The post now says the 13% rate is for the end of 2025 and adds the ruling. The 13% figure is unchanged.
Clarified
How AI Is Reshaping Small Businesses (Not Just Big Tech): An earlier version of this post said most AI tools are subscription-based. We found no source for "most." The post now says many AI tools can be used free or for a low monthly subscription fee, as a November 2025 OECD report on small firms and generative AI puts it, and cites that report.
Clarified
SAVE Act: Costs, Burdens, and Benefits Compared: An earlier version of this post said that even sources often cited to argue that noncitizen voting is a problem tend to imply a low incidence at national scale, and that worker churn increases error rates. Neither statement had a source, and the Heritage Foundation, the example given, reads its map the other way. The post now says such sources do not measure how often it happens, gives Heritage's own reading, and says churn raises the risk of errors, citing a 2024 Votebeat report on Pennsylvania.
Clarified
SAVE Act: Costs, Burdens, and Benefits Compared: The statement that audits, prosecutions and post-election reviews generally find very low levels of documented noncitizen voting had no source. Its wording is unchanged, and the post now adds a sentence with two examples: Georgia's 2024 citizenship audit found 20 noncitizens among 8.2 million registered voters, as reported by the Associated Press, and in the Kansas case the federal courts found at most 67 noncitizens registered or attempted to register over 19 years.
Clarified
Corrections Reform: What Works, What Fails, and What It Costs: An earlier version of this post said, without a source, that many returns to custody are technical violations and destabilization cascades, not new violent crimes, and that substance use disorder and serious mental illness are overrepresented in incarcerated populations. The sources now cited support narrower statements, and the post now makes those. The CSG Justice Center's Confined and Costly finds that technical violations of probation or parole, such as missing appointments or failing drug tests, account for nearly 1 in 4 state prison admissions; it does not say whether the other returns involve violent crimes. Two Bureau of Justice Statistics reports find drug dependence or abuse, and serious psychological distress, far more common among prisoners and jail inmates than in the general population; serious psychological distress is a survey measure, not a diagnosis of serious mental illness.
Corrected
Corrections Reform: What Works, What Fails, and What It Costs: An earlier version of this post said, in the reentry section and the "What works" table, that the first weeks after release are a high-risk window, and framed reentry around "predictable failure points" and "the common breakdowns immediately after release." We found no source for these, and the cited sources do not make them. The post now omits the high-risk window statement and describes reentry support as addressing practical obstacles after release.
Clarified
Why Prescription Drugs Cost More in the U.S. (and What Actually Lowers Prices): Two passages were broader than their sources. The post said Medicare Part D was created in 2006; the law was enacted earlier, and coverage began January 1, 2006, so the post now says Part D began in 2006. The descriptions of Germany, the U.K. and Canada said Germany uses reference pricing when negotiations fail, that a U.K. manufacturer must meet a price threshold, and that Canada uses reference pricing to keep prices near peer norms. The post now describes each system as the Commonwealth Fund's country profiles do: Germany negotiates prices with health funds after a new drug's added benefit is assessed, the U.K.'s NICE assesses cost-effectiveness, and Canada's provinces negotiate jointly while a federal board regulates the introductory prices of patented medicines.
Corrected
Why Prescription Drugs Cost More in the U.S. (and What Actually Lowers Prices): An earlier version of this post said Germany uses reference pricing when price negotiations fail. When talks fail, an arbitration board sets the price, using European price levels as its standard; reference prices apply to new drugs found to add no benefit. The post now says so, citing Germany's Federal Joint Committee.
All corrections and clarifications by month
2026
- October 2026, 4 entries
- September 2026, 58 entries