Our about page says corrections are timestamped and published, and that we do not quietly edit errors out of existence. This is that page. It exists because a promise you cannot check is not a standard.
Most publications correct what readers catch. Almost all of the entries below were found by auditing pages that had already passed every automated check we had — which is how we learned that those checks were comparing the site to itself and never to the source. The gates that now exist came out of these errors.
2026-07-22
Wildfire and cigarette smoke profile counts — The Smoke That Stays
Was: "among them are 26 wildfire profiles and 8 for cigarette smoke." Now: 98 wildfire profiles and 8 for cigarette smoke, with the gas-phase split (26 and 1) stated separately. Both original figures were real, but 26 counted only gas-phase wildfire profiles while 8 counted all cigarette profiles — two different denominators on either side of one comparison, which made the two sources look closer in scale than they are. The scorer now records both counts so the claim can be checked.
Firefighter gear storage — The Smoke That Stays
Was: an opening scene in which a firefighter hangs turnout gear in a mudroom "because that's where it belongs," described as properly stored by department standard. Now: gear stays at the hall, which is actual practice for career firefighters; what comes home is absorbed through the skin and on the station uniform. Volunteers, who commonly do store gear at home, are named separately. The original was simply wrong about the fire service, and it was wrong in a piece written for firefighters.
Bedroom breathing-zone claim — The Smoke That Stays
Was: a "20 to 30 centimetre breathing zone" and a claim about poor bedroom ventilation, attributed to a study cited on another page of this site. Now: the paper (Molinier et al. 2024) says neither of those things. Replaced with what it does report — high emissions of nearly 100 VOCs in the bedroom during sleep compared with other rooms of the same home — and cited directly rather than by reference to our own page.
2026-07-21
Child and adult exposure multipliers — The Smoke That Stays
Was: "toddlers received approximately 6.8 times and infants approximately 24 times the adult-equivalent dermal dose." Now: 6.8 times higher in toddlers and 24 times higher in adults, measured against what a passive smoker would inhale, with the children's figure being oral rather than dermal exposure. Three errors in one sentence: the 24× figure belonged to adults and was reassigned to infants, inventing an infant multiplier that appears nowhere in the paper; the baseline was wrong; and the route was wrong. It sat in the section addressed to parents of babies.
PAH persistence attributed to cotton — The Smoke That Stays
Was: "cotton-bound surface concentrations elevated above background for approximately 40 days," in six places. Now: the ~40 days is reported across all three materials the study tested, and its own decay constants make cotton the fastest to clear. The claim pointed the opposite way from the paper, in the direction that happened to suit our argument about bedding.
Surface cleaning versus air cleaners — The Smoke That Stays
Was: surface cleaning worked "more effectively and more persistently than portable air cleaners and more persistently than window opening." Now: more effectively than air cleaners, and more persistently than window opening — which is what the paper says. A comparison had been grafted onto the wrong comparator.
CertiPUR-US formaldehyde limit — Formaldehyde
Was: "requires certified foam to emit less than 0.1 ppm formaldehyde (small-chamber test)," in two places. Now: CertiPUR-US certifies foam as made without formaldehyde — a content criterion, not an emissions limit — and caps total VOCs at 0.5 ppm. There is no formaldehyde-specific emission threshold in the program. The figure did not exist. Our own CertiPUR explainer had it right the whole time.
A citation pointing at the wrong paper — Formaldehyde
Was: Trocho et al. 1998 on aspartame, carrying PMID 11991085. Now: PMID 9714421. The original PMID resolves to a 2002 paper on the cytotoxic effects of methanol and formate on rat thymocytes — an unrelated study. A wrong locator is worse than a missing one: it sends a reader confidently to the wrong place.
Two claims labelled peer-reviewed that named no study — Formaldehyde
Was: "A 2011 study measured 1,660 ppm formaldehyde in a popular hair styling gel," and a houseplant claim beginning "subsequent research has consistently found." Both carried a peer-reviewed tag. Now: the first is rebuilt on Lefebvre et al. 2012, which actually measured this and whose result runs against the original framing; the unmeasured deposition step is retagged as inferred. The second names Cummings and Waring 2020. No study matching the 1,660 ppm figure could be found.
Misattributed authorship — The Smoke That Stays
Was: "Li, Farmer, Goldstein et al. 2023," in twelve places. Now: Li et al. 2023. Goldstein is not an author of that paper. He is an author on three others cited in the same piece, which is likely how it happened.
Citation years and identifiers — sitewide
An audit of every peer-reviewed citation against CrossRef and Europe PMC corrected 47 mismatched records: identifiers pointing at unrelated papers, descriptions used in place of published titles, wrong first authors, and wrong years. A further 131 citations whose only locator was a search-engine URL were resolved to DOIs. This work is ongoing; 32 mismatches and 72 citations without a primary identifier remain, and are listed in our audit reports rather than hidden.
How these were found
Not by readers. By building checks that compare our published claims to the sources they cite, then running them over everything already published. Every error above passed our previous checks, because those checks verified that the site was internally consistent — that the markup parsed and the links resolved — and never asked whether a cited paper existed or said what we claimed.
Three checks now run on every change: citations must resolve to a real record whose title, author and year match what the page says; claims must name a source a reader can check; and any claim carrying a number is pinned to the source's verbatim sentence, with the number required to appear in it. A commit that adds a citation we cannot verify is refused.
We expect to add to this page. A publication that never corrects anything is either not checking or not saying.