Early on, a keyword that looked like one of the biggest numbers in the whole plan turned out to be an entire product family, not one keyword. The first pull said 165,000 a month. The real figure is nearer 3,700–18,000. A separate finding turned out to be an artifact too. It said that searches for a smoke air purifier peak in the heating season, and it rested on a seasonality array carrying only four distinct values. The five-year trend curve peaks in fire season instead. Roughly twenty claims like these got caught and retracted before anything shipped. This page is what happens when one gets caught after.
What gets corrected
This log corrects anything published on the site that turns out wrong: a number, a threshold, an attribution, a date. A reading, a quote, or a stated figure that disagrees with a reader’s own measurement is worth reporting. Readers running their own gear — PurpleAir, AirGradient, a Home Assistant dashboard — catch most of these. That habit shows up in roughly 150 posts of the material reviewed.
It’s exactly the kind of thing this log exists to record. They compare our number against theirs. We’d rather they find a dated correction than a quiet edit.
How a correction is marked
Every correction gets a date. It stays visible. Nothing gets edited away quietly, and nothing gets folded into the original text as if it always read that way. A corrected page carries a dated link back to this log. A reader lands on the history, not a clean surface pretending nothing changed.
The standing rule
The standing rule: an admitted gap beats a confident invention. Two live examples show it. Three states — Washington, Minnesota, Wisconsin — have a documented daycare air-quality threshold in the material reviewed, and no other state gets named. A city’s monitoring station, and its distance from the metro center, stay unlabeled until the data pipeline confirms them. That’s preferred over shipping a plausible-sounding number nobody checked.
A fourth, Utah, has no published rule in anything found so far, so none is stated for it. Both are gaps said out loud, not numbers guessed at.
Report one
If a number here doesn’t match your own reading, send it to us. If a quote looks wrong or a threshold reads stale, send that too. The form takes an email address only if you want an answer — reporting a wrong number shouldn’t require identifying yourself. This log is where it gets fixed in public, not quietly.
The log is below. Read how we measure for the mechanics behind the numbers, or what this project is and isn’t for the rest.
Logged corrections
Every correction BreatheCheck has made, dated, with the page it affected and what it says now. Entries are added, never edited away, and each corrected page carries a dated link back to its entry here.
15 September 2026 — the Phoenix history counted the hours after an instrument fault. Phoenix showed a daily maximum of 496 for 20 August. The readings that morning went 49, 1210, 1128, 496, 176, 104, 69 while Scottsdale, in the same valley, read 38. The first two were already held back as above the top of the scale (the correction of 6 September). The next three were the same fault: the reported number is smoothed across recent hours, so one bad hour drags the following ones with it, and anything under 500 was published. Readings in the hours after an off-scale one are now held back too, until the series returns to within reach of its last good value, for at most twelve hours. Phoenix is the only metro affected — six readings, and one day that no longer counts as above AQI 100.
15 September 2026 — the station section promised a table that did not exist. Methodology said every city page carried a station line with the monitor’s distance and a nearby-sensor count, and that a table of them would ship “once station coverage is verified”. No page rendered that line and there was no table. Thirteen city pages did print a distance, in a separate block — values typed in before launch, which nothing checked; the other 105 metros had none. Checked against AirNow’s own coordinates, only Salt Lake City’s was right. Denver’s said 1.2 miles and is 11.4; New York’s said 1.8 and is 9.4; Phoenix, Atlanta and Minneapolis were out by two to four miles; the rest by under two. Distances are now computed every hour from the coordinates AirNow returns with each reading, the station line renders on the pages whose text refers to it, and the table is on the methodology page. The sensor count stays off while the sensor feed is down, rather than showing zero.
14 September 2026 — the city index still said Kansas City had no page. Air quality by city opened with “Kansas City missed this list by 20 searches a month”. Kansas City has had a page since the index moved to the lower, measurement-based bar — the error logged below on 16 August. That correction rewrote the body of the page but not its opening line, which is drawn from a separate field and also served as the page’s description in structured data, so the same false sentence stayed in the two places a search engine reads first. The same structured data listed 28 cities while the page linked 69. The opening line now describes the page without naming a metro, and the list in the structured data is built from the same set of cities the visible table is, so the two cannot disagree again.
6 September 2026 — the Phoenix history table published readings above the top of the scale. Phoenix rendered a daily-maximum table whose figures included 2031, 1478, 1210 and five more above AQI 500 — the value at which the EPA index ends. They were never plausible: at 06:07 on 24 August the Phoenix monitor reported 2031 while Scottsdale, fifteen kilometres away in the same valley, reported 44 and Tucson 41. One instrument was misreporting. The readings had been in the series since August and the bar chart had concealed them, because the bars clip at a fixed height; they became visible the day a table was added beside the chart, and they were live for about an hour. Readings above 500 are now held out of the daily maximum, and the page states how many were held back, the highest of them, and why — they are not clamped to 500, which would invent a number, and not deleted, which would hide one.
16 August 2026 — the city index described pages that exist. Air quality by city explained that Orlando, Tampa, Honolulu and Kansas City sat below the publishing floor and that “none of the three gets a page, or should read as though it does”. All four have pages, and Cleveland — described as absent for the same reason — has one too. The passage was written when a metro earned a page only by clearing 500 searches a month, and it was not revisited when a second, cheaper kind of page was introduced at a lower bar. The section has been rewritten to state both bars and what each buys a reader. No reading or threshold was affected; what was wrong was the site describing itself.