When a record is wrong, we fix it and say so. This page states the standard.
What counts as a correction
A factual error on a record: a wrong tally, a misspelled official’s name, a wrong amount, a wrong date, a wrong jurisdiction, or a status that does not match the documented outcome. Corrections are not updates: a record gaining a newly published roll call, the way Toledo’s did, is growth, and the record’s last-verified date reflects it.
The categories so far
The corrections queue has handled these classes: records created for a city already tracked under an older naming pattern (merged, with the duplicate trashed); fields wiped by a tooling error and restored from sources; a computed headline that misread a partial vote table (fixed the same day the pattern was found); approving bodies that named a department or a process in place of the body that approved; and source dates that recorded the day a page was read. Each fix follows the same path: verify against the underlying source, correct the record, and re-verify the surfaces that display it.
The log
Every correction made to a published record or figure, newest first. Each entry states what was wrong, what it says now, and how it surfaced.
September 24, 2026. The Norfolk record misdated a court ruling. It said a federal judge ruled in February 2026 that the Institute for Justice suit could proceed. That ruling, which denied the city’s motion to dismiss, came on February 6, 2025; in January 2026 the court ruled for the city, and the plaintiffs said they would appeal. The record and its source date now match the Institute for Justice release. This surfaced while the lawsuits page was brought up to date.
September 24, 2026. 18 approving bodies named a department, a process or a sentence. Richmond listed the police department and a city procurement route, Suffolk listed the city budget, and Mountain View, Santa Cruz, Staunton and Springfield listed only city government. Each record’s sources were read again. Fair Oaks Ranch now names the Fair Oaks Ranch City Council, because the city’s own presentation says the purchase was approved in the fiscal 2024-25 budget. The other 17 name no approving body, because none of their sources names one: Suffolk, Mountain View, Santa Cruz, Staunton, Springfield, Eugene, Rehoboth Beach, Glastonbury, Everett, Richmond, Oxnard, Mill Creek, Bloomington, Providence, Charleston, Cullman and the Rhode Island State Police. Suffolk and Glastonbury also moved from council vote to unknown, since their sources say only that the money went through the budget. These surfaced in a site review.
September 24, 2026. 9 records listed a council vote without naming the council. 5 sources name the body that voted, and those records now show it: the Norfolk City Council, the Wichita City Council, the Dayton City Commission, the Eastport South Manor school board and the Cleveland Metropolitan School District school board. The sources for Fort Worth, Fargo, Castle Rock and Rohnert Park name no approving body, and the only Rohnert Park council vote they report set a retention limit, so those 4 records now show the approval method as unknown. Pitt County moved the same way: its source reports a Board of Education vote on cameras at 2 schools and does not say who approved the sheriff’s contract. These surfaced in a site review.
September 24, 2026. Appleton now shows its approval date. The record named the approving body but carried no date. Its Legistar vote record belongs to the Common Council meeting of October 4, 2023, where file 23-1130, the police department’s sole source request for Flock Safety cameras, passed on a motion by Christopher Croatt seconded by Brad Firkus. The record now carries October 4, 2023 and names the body as Common Council, the name Legistar uses. This surfaced in a site review.
September 24, 2026. 3 source dates were the day the page was read. A City of Cambridge statement, a Times Republican report on Marshalltown and a City of Richmond release each carried July 16, 2026 as their date. They were published in December 2025, February 2026 and July 2025. The dates are now blank and each label names the month. These surfaced while source labels were rewritten.
September 24, 2026. Page figures that were wrong or out of date. The About page and the states page listed New Hampshire among the states with no record; New Hampshire has 2, and 47 states and DC carry at least 1. The glossary said no Flock contract exists in New Hampshire; that claim is gone, and the 3-minute retention rule now links the statute. The home page counted 48 states, a figure that included DC. The data page and this page said about a third of records lack an approval date; the share was 72 percent, and both pages now read it live. The tracker page said 29 records carry an end date; 104 did, and the page now reads the live count. The home page promised 20 FAQ questions; the FAQ has 10. The Flock explainer said every Flock contract is a lease; it now says most are subscriptions and names a city that owns its cameras. The privacy policy did not name Google Analytics; it now does. These surfaced in a review of the site’s copy.
September 23, 2026. Utah records rebuilt, approving bodies removed. 16 Utah records had a state search form as their only source. All 16 were unpublished, and 12 were republished after being rebuilt from primary documents. 9 records named an approving body that their sources did not name, and those bodies were removed. The Skamania County, Washington record now names the county as its jurisdiction. 11 records labeled as purchases under a spending threshold had no documented dollar threshold, and their approval method now reads unknown. These surfaced in a sourcing audit.
August 5, 2026. The site total moved from 266 records to 264. Two cities were tracked twice under different slug patterns: Los Angeles at /city/los-angeles-ca/ and again at /city/ca-los-angeles/, and Bandera at /city/bandera-tx/ and again at /city/tx-bandera/. Reading both pairs in full showed each pair documented the same contract, so the thinner record in each pair was merged into the richer one and trashed. Both retired paths now return a 301 to the record that survived. The duplicates were found by a scan for repeated city and state pairs in the JSON export.
August 5, 2026. The Los Angeles contract end date moved from June 30 to July 11, 2026. The two duplicate records disagreed, one carrying June 30 and the other July 11. Both cited the same TechCrunch report of July 13, 2026, which states the three-year agreement would expire when it ended that Saturday. The Saturday before July 13 was July 11. The record now carries July 11 with that reasoning written into the summary. The older figure had been labelled on the record as a reported June expiry rather than a confirmed date.
August 5, 2026. A West Hollywood source was removed from the Los Angeles record. While merging the duplicate pair, one of the sources on the thinner Los Angeles record turned out to be a Beverly Press article about West Hollywood keeping its cameras. It documented a different city and was not carried over to the surviving record.
August 5, 2026. Bandera’s camera count moved from 8 to 5. The Texas deep dive described an 8-camera program. The record documents a $17,000 deal covering 5 cameras, mostly paid by a $14,000 Motor Vehicle Crime Prevention Authority grant. The documented purchase governs, so the post was corrected to match the record rather than the other way round.
August 5, 2026. The Texas post moved from 18 records to 19, and its URL changed. Texas gained a record after the post was published, so the title and body said 19 while the slug still said 18. The post now lives at /texas-and-flock-2-extremes/ with a 301 from the old address, and the slug carries no count so it cannot drift again. Every internal link to the old address was repointed.
Two of these came from our own scans rather than from a reader. That is the intended order: the checks should find the error before anyone else has to.
How to report an error
Use the correction form at the foot of this page, with the record URL and the source that contradicts it. We review each correction within 72 hours; most are read the same day. A correction with a source outranks the record as written; we check the source, fix what it proves, and note the change. The faster we hear, the faster the record is right.
What to send
The record URL, the fact that is wrong, and a link to the document that shows it. That third part carries the weight. A message saying a camera count is wrong starts a records request; a message with a link to the purchase order changes the record the same day. Screenshots help when a page has since changed, but a live link to the source is better.
What happens next
The claim gets checked against the sources already on the record, then against the document you sent. If they conflict, the underlying government document wins over reporting about it, and a newer document wins over an older one. If the correction holds, the record changes, the source list grows, and the last-verified date moves. Every page that counts that record recomputes on the next save, so the tracker, states, and officials pages move together.
What a correction looks like on a record
Corrections stay visible rather than silent. When sources disagree, the record says so in plain words: the Alameda County record carries a 3 to 2 roll call alongside a broadcast report describing the vote as 4 to 3, and states the discrepancy instead of picking quietly. The Norfolk record explains why it uses a 172-camera deployment figure when the vendor portal lists 175.
The standing policy
An empty field beats a wrong one. If a document does not state a figure, the record leaves it blank and the summary says what is missing, which is why 73 percent of records carry no approval date. A skip with a stated reason beats a stretched record. Tallies without names stay tallies until a journal publishes the names, and termination votes never enter a record’s roll call, which documents how a contract was approved.