Friday, August 21, 2026
Rewritten from 136 words to 641 around Chanthorn and colleagues, Scientific Reports 2019, which mapped the seed-dispersal network of a 30-hectare plot in central Thailand and simulated removing the large frugivores: trees dependent on those animals hold nearly a third of above-ground carbon, and complete defaunation reduces stored carbon by 2.4 to 3.0 percent, exceeding one percent once defaunation passes forty percent. The original claimed the canopy 'collapses' and cited nothing.
We overstated it, and in the direction that makes a story easier to tell. Three percent is not a collapse — it is a substitution, an old dense forest quietly becoming a young light one with the same outline. That is harder to campaign about and more alarming in one specific way, because nothing visible happens and so nothing triggers a response. Correcting an exaggeration matters as much as correcting an error; a publication that only ever revises upward is not measuring anything.
All 2 versions of this piece →
Rewritten from 138 words to 620 around Feely and colleagues, Global Biogeochemical Cycles 2012: the Pacific aragonite and calcite saturation horizons migrating upward at about 1 to 2 metres a year, and a rise of more than 100 metres in the California Current between the 1994 WOCE survey and the repeat occupation. The original asserted the direction with no rate, no region and no source.
A claim about a moving threshold is only useful with a rate and a place attached, and the regional figure turns out to be the interesting one: on this coast circulation does much of the work, which is why acidification damaged Pacific Northwest hatcheries before the projections expected it. The piece now also separates the chemical threshold from the biological outcome instead of treating crossing the horizon as an obituary.
All 2 versions of this piece →
Rewritten from 141 words to 662. The original attributed heat-dome intensity to 'stratospheric inversion' — the wrong layer of the atmosphere, and not a term used for this. The cap is a subsidence inversion in the troposphere, made by adiabatic warming as air descends inside a blocking anticyclone. Built out around the June 2021 Pacific Northwest event: a 500 hPa height of about 5,980 m with no precedent for the region since 1948, and the Earth System Dynamics attribution analysis putting it at 1-in-1000 years now, at least 150 times rarer without human warming, and about 2 °C hotter than the same-rarity event in 1850–1900.
We named the wrong layer of the atmosphere as the mechanism, which is the kind of error that quietly teaches a reader something false. The corrected version also earns its local paragraph honestly: a blocking ridge suppresses the pressure gradient that pulls the Delta breeze inland, so the nights stop cooling — and overnight temperature is what the heat-mortality evidence actually turns on. It also states plainly what is unsettled, rather than implying blocking frequency is a closed question.
All 2 versions of this piece →
Rewritten from 128 words to 641 around the MESA Air and Lung cohort in JAMA, August 2019: 7,071 adults, up to five CT scans each, a median ten years of follow-up. The original cited 'recent histological analysis of urban residents' with no study behind it. The piece now carries the measured numbers — 0.58 percentage points of emphysema per decade from age alone, plus 0.11 per 2 µg/m³ of PM2.5 and 0.13 per 3 ppb of ozone — and the finding that reframes it: as particulate concentrations fell, ozone was still associated with progression and PM2.5 was not.
The original made a physiological claim with no evidence attached, and the real evidence points somewhere more useful and more local: the pollutant still taking lung tissue at today's concentrations is ozone, and the Sacramento Valley is one of the country's hardest ozone problems. A piece that names the right pollutant for the reader's own air is worth more than one that is vaguely right about pollution in general.
All 2 versions of this piece →
Rewritten from 138 words to 658 around the EPA's technical assistance document for reporting daily air quality. The original named a 'Variance Protocol' that does not exist — the term was ours — and said real-time AQI is a 24-hour average hiding acute spikes. The document sets out the actual PM NowCast: a twelve-hour window that shortens toward three hours as air becomes variable, introduced in 2013 to be MORE responsive during fires. A correction sits at the top and the figure shows the three windows.
We published an invented technical term as an established government mechanism, and built an accusation on it. That is the most serious thing on this site and it stood for eleven weeks. The three points in the original that were true — the daily figure really is a 24-hour average, the NowCast goes quiet on missing data, and a monitor measures where the monitor is — are kept and sourced, so the correction does not throw away the reporting with the error.
All 2 versions of this piece →
Rewritten from 136 words to 585, around the USDA's own long-run measurements: 14.9 tons per acre per year in 1930, 7.8 in 1982, 6.3 in 1992, 4.63 in 2017. The original claim — that topsoil is eroding 10 to 40 times faster than it forms and accelerating — was unsourced and points the opposite way to the federal record. A correction now sits at the top of the piece, and the figure shows the four measurements.
We published a number we could not source, in a direction the evidence does not support, and left it up for eleven weeks. Correcting it in public with the primary series attached is the only version of this piece worth having — and the harder question it leaves open, about what remains in the soil that stayed, is stated as unanswered rather than filled with another ratio.
All 2 versions of this piece →
Thursday, August 20, 2026
The opening line was rewritten. It was one 21-word sentence reading at grade 14.8; it is now two short ones. The five markers and the single cause are also stated as figures in the text, so the numbers printed beside them appear where a reader can check them.
This desk promises plain language, and the hardest sentence on the page was the first one a reader meets. The publishing gate flagged it and we shipped anyway; publishing the fix the same hour, with the reason, is the correction being made in public rather than quietly.
All 2 versions of this piece →
The review opens, holding the first finding.
A desk that publishes daily needs one place a reader can go instead of reading everything in order. Opening it now, with one finding in it, means the document's history starts at the beginning rather than being backfilled later from memory.
All 1 version of this piece →
First published.
The desk opens on the only case it can check completely: its own site, measured rather than described. It sets the method every later finding has to meet — instruments, numbers anyone can reproduce, and a stated limit.
All 1 version of this piece →
Tuesday, May 19, 2026
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →
First published.
The version this publication started from. Recorded when the ledger was built, on 20 August 2026, so the archive has a floor rather than appearing unedited.
All 1 version of this piece →