Corrections Policy

Change: The piece gave Flock's read volume as "more than 20 billion plate reads a month". The article it cites, in The Drive, quotes a Flock spokesperson saying the system performs 20 billion reads a month. The piece now gives the figure as its source does.

Reason: A figure attributed to a source has to match the source, and the words "more than" were the piece's own addition.

Change: The section on the three ads said their sponsorship was different. The sources name who sponsored two of the three: the Rogers campaign with the National Republican Senatorial Committee (the Rogers for Senate page) and a super PAC backing Rogers (Mediaite). The Gateway Pundit and WorldNetDaily, the two reports on the Ogonowski spot, do not say who sponsored it. The line now says the length, wording, and evidentiary claims are different and that neither report names the Ogonowski spot's sponsor.

Reason: Saying three ads have different sponsors is a claim about who paid for each, and the sources name the sponsors of only two of them. The distinction between the ads rests on their content, which the sources do show, and is unchanged.

Change: A section headed "The plaintext vulnerability" said that Canary Mission's database is stored in a form that makes transfer frictionless and that the databases behind the door-knock pipeline "are readable, copyable, and portable because they were never encrypted". No cited source says how Canary Mission stores its data, and the site publishes its dossiers, so the material is free to read without any question of encryption arising. The section now says that: the pipeline works because nothing in it was ever made expensive to open. The pattern is still named and the argument for retroactive encryption is unchanged. The same claim was corrected in the Flock piece as correction #16 and in The Archive Is Already Built as correction #22.

Reason: The piece asserted a storage format for a named organization that no source establishes, when the fact that carries the argument is simpler and documented: the dossiers are published.

Change: Correction #17 removed the words "subject to court orders" from the piece's description of Senate Bill 10, because the enrolled text of the bill does not say that. Removing the phrase was right, but it left the piece with nothing about the litigation the phrase gestured at, and that litigation is the reason readers ask whether the law is in force. The piece now says what the record shows: Texas families represented by the ACLU of Texas, the ACLU, Americans United for Separation of Church and State and the Freedom From Religion Foundation filed three suits against SB 10 beginning in July 2025; a federal district judge blocked eleven districts from displaying the posters in August 2025 and a second judge ordered fourteen more districts to remove theirs by December 1, 2025; and on April 21, 2026 the full Fifth Circuit upheld SB 10 by 9 to 8 in Nathan v. Alamo Heights Independent School District and vacated the first injunction, and it later reversed the second in light of that ruling. On August 17, 2026 the plaintiffs asked the Supreme Court to take up that case and a companion case, Cribbs Ringer v. Comal Independent School District. The ACLU of Texas case page, a Courthouse News report of the ruling and SCOTUSblog's report of the petition are added as sources.

Reason: Removing an unsupported phrase is only half a fix when the subject it pointed at is real, current and material to the reader.

Change: The essay said that Canary Mission's database "was stored in plaintext" and that Flock Safety's movement records are held in plaintext, and it used both as examples of the pattern plaintext liability. Neither claim was sourced. Flock's evidence policy says the records are encrypted at rest with AES-256, which is the opposite, and the site's own lexicon entry lists an institution that encrypted at rest as a false positive for the pattern. Canary Mission publishes its dossiers, so nothing had to be breached or decrypted for them to be read. The section now says what each case actually shows: Canary Mission drove the cost of reading to zero by publishing, and Flock encrypts at rest, but the customer that owns the records can still search and export them and share access with other agencies, and Flock says it redirects legal process to that customer, so encryption at rest leaves the cost of reading where it was for everyone the customer lets in. The essay's demand is restated accordingly, as a demand about what a read costs rather than about storage format alone, and two later passages that assumed the plaintext framing were rewritten to match. Flock's evidence policy and FAQ are added as sources. The same claim was corrected in the Flock piece as correction #16.

Reason: A claim about how a named company stores its data has to match what that company publishes, and the essay was contradicting the site's own definition of the pattern it was invoking.

Change: The piece's description and one line said all four lawmakers cited the New England Journal of Medicine, and its search title said they cited "their own source". FactCheck.org and PolitiFact show that Rep. Lisa McClain named the journal, on Fox Business on August 11; that Sens. Paul and Johnson's press release elevated the 82 percent figure from a 2022 letter by Dr. James Thorp; that Sen. Cruz repeated it on X and his podcast on August 12; and that Sen. Johnson cited a 2021 study. The description, the search title and the line now say the four repeated the figure and that McClain named the journal.

Reason: Saying four people named the same source is a claim about each of them, and only one did. The piece's argument, that the figure is not in the study, is unchanged.

Change: The piece called Fauci's two statements "one message, not two" and said they came "four and a half years" before anyone needed only the first half. FactCheck.org describes a text thread: on January 25, 2021 he wrote that fever after the second dose "theoretically could be associated with miscarriage in the 1st trimester", and on January 26 he wrote that more than 10,000 pregnant women had been vaccinated and "no issues have arisen". The piece now describes two texts on two days in one thread, quotes both, and says more than five years, counting from the January 2021 texts to their release on August 10, 2026.

Reason: The section argues that the concern and its answer belong together, and that depends on how the exchange ran. FactCheck.org, cited by the piece, gives the dates and the wording.

Change: The piece said the Columbia Journalism Review describes the Detroit News as a traditionally conservative Republican newspaper. That sentence is from an essay CJR published on November 1, 2024, written by Gary Miles, the editor and publisher of the Detroit News. The piece now names him and the essay, quotes his sentence, and lists the essay in its sources with his role.

Reason: Who says a thing is part of the claim. This is the paper's own account of itself, published by CJR, not CJR's judgment of the paper, and a piece about a label placed on that paper should get the difference right.

Change: The piece said that Brad Parscale's "own contract language, quoted by Axios back at the September 2025 filing, committed to building sites specifically to shape how AI systems summarize conversations about Israel". The document is the services agreement filed with the registration, and its pages are marked "CLOCK TOWER DRAFT" and dated August 27, 2025. Among the services it lists is "Deployment of websites and content to deliver GPT framing results on GPT conversations." Axios quoted that line and described it as "seemingly to create new data for AI models to train on". The line does not mention summaries or Israel. The piece now says what the draft lists and attributes the reading of it to Axios and Drop Site News. The opening lines also said the ten sites had "a combined audience smaller than a mid-sized newsletter". No source gives that comparison. Drop Site News reports that most of the sites average a few hundred unique visitors each month, and the piece now says that.

Reason: A filing marked as a draft should be described as one, and a claim about what a contract required should quote what the document says. A comparison that no source supports cannot be checked by a reader.

Change: The comparison table said Senate Bill 10 requires a Ten Commandments poster or framed copy in each classroom "subject to court orders". The enrolled text of the bill does not mention court orders. Its display requirement is subject to a different provision: a school that lacks a poster in each classroom must accept privately donated posters and display them, and may, but is not required to, buy posters with district funds (Education Code § 1.0041(a), (d) and (e)). The row now says that. The piece also did not say that Senate Bill 11 provides for the attorney general to defend districts that are sued. On request from a board, the attorney general must defend the district in a cause of action arising out of its adoption of a policy. If the attorney general does, the state is liable for the expenses, costs, judgments or settlements of the claims arising out of the representation, and the state may not be liable for those costs for a district the attorney general is not representing (§ 25.0823(e) and (f)). The piece now says so.

Reason: A description of what a statute requires should match its text, and who pays for the defense of a lawsuit is part of what the statute does.

Change: The last section of the piece said Flock stores its movement records "in plaintext", named the pattern plaintext liability, and proposed encrypting the records at rest as the remedy. No cited source says the records are stored unencrypted. Flock's evidence policy says all stored data, including footage and metadata, is encrypted at rest with AES-256, and its privacy policy says customer data is stored using AES256 encryption. The section and the piece's two pattern tags for it (plaintext liability and retroactive encryption) were removed, and the piece now reports what Flock's evidence policy says. The piece also said that the Associated Press "later found" that Border Patrol had access through Flock to at least 1,600 readers in 22 states, "while local agencies also performed searches on CBP's behalf". The AP article, published November 20, 2025, says that through Flock alone Border Patrol for a time had access to at least 1,600 license plate readers across 22 states, and that some counties reported looking up plates on behalf of Customs and Border Protection, even in California and Illinois, according to an AP analysis of police disclosures. The piece now says that, adds Flock's statement to the AP that it had "for now" paused its pilot programs with CBP and Homeland Security Investigations, and links to the AP article.

Reason: A claim about how a company stores data should match what the company's own policies say, and they say the opposite of what the section claimed. A finding attributed to a news organization should say what that report says.

Change: The page's summary said most of the AI systems asked about the disclosure did not mention it. Drop Site News reported that of five chatbots it asked about one of the sites, three (Claude, ChatGPT and Gemini) mentioned the disclosure and two (Perplexity and Microsoft Copilot) did not. The summary now says so, as the body of the piece already did. The piece also now says that the foreign-agent filing describes the campaign as a nationwide campaign in the United States to combat antisemitism, and the filing's Exhibits A and B are now listed among its sources.

Reason: A summary that reverses a count misstates the finding it summarizes, and the purpose the filing gives for the work is part of what the filing says.

Change: The piece gave Flock's default retention period as thirty days. Flock's privacy policy, last updated in August 2025, says thirty days, but its LPR policy (updated June 30, 2026), its evidence policy and a June 2026 blog post say the default is seven days, so the piece now says seven and notes the difference. Details that the pages the piece cites do not contain were removed or reworded to match them: a figure for frames captured per vehicle, a quotation about an "interconnected system", a claim that the cameras do not classify race or gender, past federal pilots with the FBI, ATF and NCIS, an April 2026 date for Audit Assistance, and descriptions of the search logs as permanent and immutable. The Orange City, Florida case now follows the Institute for Justice's account (more than 100 searches over seven months, arrested and charged in 2025) in place of search counts that page does not give. The Johnson County, Texas abortion-search passage now says what The Record reports, and no longer gives a camera count, a logged reason, prosecutors' involvement or the absence of charges that its page does not contain, or says Illinois officials found a violation. Statements about Illinois audits, a March 2026 acknowledgement by Flock, other courts and a Flock legal argument had no cited source and were removed or replaced with what Flock's own pages say. The Oshkosh vote is described as unanimous, as WBAY reports. A sentence about the reasoning of Chatrie v. United States now quotes the opinion, and the statements about the Columbia, Missouri case and the Norfolk appeal now cite sources.

Reason: A retention period is a specific claim about a company's practice, and its own current policies say seven days. A detail that no cited page contains cannot be checked by a reader, so it was removed or replaced with what the cited page says.

Change: The piece listed Friendswood ISD among the districts that voted yes on Senate Bill 11. The only SB 11 document attached to the district's February 9, 2026 agenda is a resolution titled "Declining to Adopt a Policy Under SB11 Regarding a Period of Prayer and Reading of Religious Texts", and the draft minutes record the board approving "the resolution related to SB11" by 5 to 0, with one abstention and one absence. Four districts declined (Dallas, Killeen, Dripping Springs and Friendswood) and two voted to adopt (Boerne and Santa Fe), and the two sections are retitled to match. The piece also said that in Kennedy v. Bremerton School District the Court emphasized that schools cannot put students in the position of choosing between joining a religious exercise and publicly rejecting it. That language comes from Lee v. Weisman (1992), which Justice Sotomayor's dissent in Kennedy quotes. The Kennedy majority protected a coach's quiet prayer and wrote that the district's problem was a lack of evidence that coercion actually occurred. The Dallas ISD row credited the written resolution with a statement about continuing existing protections. That statement came from a trustee in the meeting, and the resolution says only that the board will not adopt a policy. The piece now also says that a person whose consent is revoked remains bound by the waiver (Education Code § 25.0823(c)(2)), and its link to the enrolled text of SB 11 now points to the enrolled version, not the committee-report version, whose operative text is the same.

Reason: A district's vote is the fact the section reports, and a quotation credited to the wrong opinion, or a statement credited to a document that does not contain it, misstates the record. That the waiver survives revocation bears directly on the consent the piece describes.

Change: The numbers list said "over 15 million YouTube subscribers" on her "Songs for Littles" channel. Her YouTube channel is named "Ms Rachel - Toddler Learning Videos" (@msrachel) and showed 21.3 million subscribers on September 19, 2026. The list now names that channel and says more than 21 million.

Reason: A channel name and a subscriber count are specific factual claims, and the count was several million behind the channel it describes.

Change: The piece described the 127 people behind the 82 percent figure as a separate follow-up group who "were still pregnant", and its arithmetic line read as 827 divided by 127. The 127 are what is left of the 827 completed pregnancies after the 700 in which the vaccine was given in the third trimester are set aside: early-vaccinated women whose pregnancies had already ended by the study cutoff. Most of the other early-vaccinated participants were still pregnant and were not counted. The 82 percent figure is the 104 miscarriages divided by those 127. The piece now says so, and it now says that the study authors later acknowledged that 827 was not a useful denominator.

Reason: A denominator is the whole argument of a piece about a statistic, and describing it wrongly is the kind of precision defect this site exists to catch. PolitiFact, cited by the piece, describes the 127 and the authors' later acknowledgement.

Change: The piece said Ms. Rachel launched the petition to close the Dilley Immigration Processing Center and delivered its signatures to Washington, and it gave the count as both 324,000 and 350,000+. She helped launch the petition, whose first signers included dozens of artists and doctors, and she promoted it. It had more than 324,000 signatures in June and has since passed 350,000. What she carried to Congress in June was letters from detained children. The introduction also counted that petition as part of the movement asking her to run for president, and the numbers list counted its signatures as support for her, but it is a petition about the detention center. Captions, image descriptions and the lines introducing the screenshots and posts claimed more than they show: that the movement was not astroturfing or a PAC, that the posters were parents or regular people, that nobody is coordinating the posts, that the posters are not bots and were not prompted or paid, that none of the accounts carry a blue check, that the posters have no audience, and that every result repeats the slogan or the year. They now describe what each screenshot shows, including a verified account posting against her and several posts from her own account, and the summary no longer calls the movement organic. An Instagram reel was also captioned as her thank-you when the Dilley petition passed 350,000 signatures. It was posted on January 20, more than two months before that petition was created, together with the youth center at Aida refugee camp near Bethlehem, and it thanks the people who signed a petition from Aida Camp against an Israeli order to demolish the football pitch there. The caption and image description now say so.

Reason: Who started a petition and what reached Congress are specific factual claims, and a screenshot cannot show that no one coordinated a post. A caption that ties a source to the wrong campaign also misstates what that source shows.

Change: The source URL for Ohio Revised Code § 2151.421 (codes.ohio.gov/ohio-revised-code/section/2151.421) has been intermittently returning 502 Bad Gateway. Replaced with the Justia mirror URL (law.justia.com/codes/ohio/title-21/chapter-2151/section-2151-421/), which hosts the same statutory text and is consistently accessible.

Reason: A source URL that returns 502 cannot serve its purpose. The Justia mirror provides the same statutory text and is a stable legal reference site.

Change: The examples field attributed the 37% manipulation rate and 14x engagement figure to "De Freitas et al., Journal of Consumer Research, 2025" (the "AI Companions Reduce Loneliness" paper). Those statistics come from a different paper by the same lead author: "Emotional Manipulation by AI Companions" (arXiv 2508.19258, 2025 working paper). Corrected the attribution and updated the source URL. Also corrected the description: the 37% figure covers all six manipulation tactic categories, not just guilt/FOMO specifically.

Reason: Attributing statistics to the wrong paper is a source-claim mismatch. The JCR paper studies loneliness reduction; the exit-flow manipulation statistics come from the arXiv working paper.

Change: The seed case excerpt for the UK patriots AI image said it was "shared 2 million times." The source says it was "viewed almost 2 million times" and reshared 16,000 times. Changed "shared" to "viewed" in both the excerpt and the hero text.

Reason: Views and shares are different metrics. The actual share count was 16,000, not 2 million. Conflating views with shares overstates the propagation.

Change: The page said Ghislaine Maxwell was convicted on "six counts." She was convicted on five of six counts, acquitted of enticing a minor to travel. Corrected to "five of six counts" with the acquittal noted.

Reason: A conviction count is a specific factual claim. Overstating the conviction count by one is the kind of precision defect this site exists to catch.

Change: The H.R.4405 entry said "Senate unanimous same day; signed by Trump the next day." The congressional record shows the House passed it Nov 18, the Senate passed it Nov 19, and Trump signed it Nov 19. Corrected to "Senate unanimous November 19; signed by Trump the same day."

Reason: The Senate did not pass it the same day as the House (Nov 18 vs Nov 19), and Trump signed it the same day the Senate passed it, not the next day.

Change: The source link for Tyler Barriss's 20-year federal sentence pointed to a January 2018 CBS article about his initial false-alarm charge, not the March 2019 sentencing article. The claim was accurate; the citation was wrong. Replaced with the correct CBS sentencing article. The Wired source title was also corrected from "Swatting Can Be Deadly. It's Also Poised to Get Worse" to the actual article title "It Started as an Online Gaming Prank. Then It Turned Deadly".

Reason: A citation that points at an article not containing the information it supports is the same defect class as a dead link. The sentence was always correct, but a reader clicking through would have found a charge article, not a sentencing article.

Change: The piece said Jake Lang served four years for an eleven-count indictment and that Donald Trump dismissed his convictions. Lang has no convictions. He spent four years in pretrial detention as the longest-held January 6 defendant still awaiting trial, and was pardoned before that trial happened. The Minneapolis passage was also re-sourced from Wikipedia onto Star Tribune and NBC News reporting, and three details in it were tightened to match what those outlets documented.

Reason: The error was inherited from Wikipedia, which describes the dismissed charges as dismissed convictions. Reporting confirms he was never tried. Saying a named living person was convicted of crimes he was only ever charged with is the kind of error this site exists to catch, so it is logged here rather than quietly amended.

Change: Verdict changed from "False" to "Institutional deception".

Reason: The previous verdict "False" was ambiguous, it was unclear whether it referred to Hillsong's denials being false or the propaganda being false. The new verdict is more precise.

Change: Verdict changed from "False" to "Institutional deception". Language softened from direct "cult" labeling to BITE model clinical framing.

Reason: Editorial alignment with clinical framing standards: use BITE model terminology instead of "cult" label. The previous verdict was ambiguous for an institutional analysis post.