find my
next bite.

What we know · Corrections and revisions

What we got wrong, and what we changed.

This page lists every change this site has made to a sentence it published, and every change it made to a sentence before publishing it. Nothing that was published is quietly rewritten: a change is a dated entry here, and the page that carries it says what changed.

49 entries · 44 corrections of an error · 5 revisions · last changed

Where a fix was made before the sentence was ever served to a reader, this page says so and does not list it as a published error. The four technology entries under What we know have been served since 20 September 2026, so a change to one of those is a change to a page a reader could already have read. A change to an unpublished draft, or to a page whose release switch is still off, is not.

Two kinds of change

Kept apart on purpose

Two kinds of change, kept apart in the words the Office for National Statistics uses in its own revisions policy: a revision is a change that improves a figure because better information arrived, and a correction of an error is a change made because something was wrong. A claim that was right when it was published and is more accurate now is a revision. A claim that was wrong is a correction.

Correction of an error
What we published was wrong. Something in the sentence did not match the source it named, or the arithmetic did not hold.
Revision
What we published was right when we published it, and later information made it more accurate or more precise. A revision is not a mistake.

The changes

Newest first

  • Why food · the sentence on water for foods other than beef

    · Correction of an error · published, then changed

    What it said
    so the factor nearly disappears for the animal foods that are not beef.
    What it says now
    Milk, eggs and chicken are only about one and a half times pulses per gram of protein, so the factor nearly disappears for those three.
    Why it changed
    The report gives the ratio for milk, eggs and chicken only. "The animal foods that are not beef" also took in pig, sheep and goat.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L14 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the short version, the sentence on land, water and feed

    · Correction of an error · published, then changed

    What it said
    Measured per unit of food, beef and lamb need roughly fifty to ninety times more land than the plant foods that would replace them, six to twenty times more water than most plant staples, and several times more feed than the animals that convert feed best.
    What it says now
    Measured per unit of food, beef and lamb need roughly fifty to ninety times more land than the plant foods that would replace them. Beef also needs six to twenty times more water than most plant staples, and several times more feed than the animals that convert feed best.
    Why it changed
    The water report gives these ratios for beef only. No lamb water ratio was retrieved.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L03 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on a tenfold energy expansion

    · Correction of an error · published, then changed

    What it said
    A tenfold expansion of energy on freed land would take about half of it, and would give up several billion tonnes of CO₂ a year that restoration would have taken up instead.
    What it says now
    On the measured footprint, a tenfold expansion of energy on freed land would take about half of it. If all of that land would otherwise have been restored, it would give up several billion tonnes of CO₂ a year that restoration would have taken up instead.
    Why it changed
    "About half" holds only on the measured footprint, and the carbon figure assumes every freed hectare would be restored. The page now says both.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L30 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sources list, the IPCC chapter on land use

    · Correction of an error · published, then changed

    What it said
    5–10 tCO₂ per hectare per year over about a century, range 1–20; land-based removal does not sequester indefinitely.
    What it says now
    5–10 tCO₂ per hectare per year over about a century, range 1–20, for reforestation and afforestation. That land-based removal does not store carbon indefinitely is from AR6 Working Group I, Chapter 5.
    Why it changed
    The permanence clause comes from a different IPCC chapter from the one the note names.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row S06 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (backend/core/why.py)

  • Why food · the sources list, the Our World in Data land article

    · Correction of an error · published, then changed

    What it said
    4.1 billion ha to 1 billion ha, −75%; about 65% of cattle grazing land is not croppable.
    What it says now
    4.1 billion ha to 1 billion ha, −75%.
    Why it changed
    The 65% was withdrawn on 21 September 2026 because no retrieved source carries it. The withdrawal reached the comparison page but not this note.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row S04 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (backend/core/why.py)

  • Why food · the sources list, the IPCC land report

    · Correction of an error · published, then changed

    What it said
    AFOLU: 81% of anthropogenic N₂O, 44% of methane, 22–23% of net GHG emissions.
    What it says now
    AFOLU, 2007–2016: 81% of anthropogenic N₂O, 44% of methane, 23% of net GHG emissions.
    Why it changed
    The linked report gives 23% for 2007–2016. The 22% came from a different IPCC report and year.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row S02 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (backend/core/why.py)

  • Why food · the sentence on land per unit of solar power

    · Correction of an error · published, then changed

    What it said
    Solar power occupies between 0.03 and 1.5 hectares per gigawatt-hour depending on the study, against 37.5 to 81 for bioenergy crops.
    What it says now
    Counting only the land a solar farm itself covers, solar power occupies between 0.03 and 1.5 hectares per gigawatt-hour depending on the study, against 37.5 to 81 for bioenergy crops, and the best measured figure, from US solar farms, is about 0.9 to 1.0.
    Why it changed
    The range counts only the land the farm covers, and the page did not say so. Its lowest figure is thirty times below the best measured one, which the page now gives.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L26 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on today's energy supply on solar

    · Correction of an error · published, then changed

    What it said
    Put today's entire global primary energy supply on solar and it takes about 1.5 million square kilometres, roughly five percent of the land that would be freed, and around nine percent of the world's cropland.
    What it says now
    Put today's entire global primary energy supply on solar at that measured figure and it takes about 1.5 million square kilometres, roughly five percent of the land that would be freed, and about 9 to 10% of the world's cropland.
    Why it changed
    The arithmetic gives 9.3 to 10.5% of cropland, and "around nine percent" rounded it down.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L27 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · what a better farm does not fix

    · Correction of an error · published, then changed

    What it said
    Yields have risen for decades, feed conversion has improved, and neither has cut the total. Total impact is area multiplied by impact per hectare, and gains per hectare tend to be spent on producing more rather than on farming less. The systems literature calls this a rebound effect; in food it is the reason sixty years of efficiency has not produced an absolute reduction.
    What it says now
    Yields have risen for decades, and feed conversion has improved. Total impact is area multiplied by impact per hectare, and gains per hectare tend to be spent on producing more rather than on farming less. The systems literature calls this a rebound effect.
    Why it changed
    No retrieved source gives the sixty years or the absence of an absolute reduction, and Our World in Data reports that the world has passed peak agricultural land. Both claims are removed.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L23 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on carbon taken up by restored land

    · Correction of an error · published, then changed

    What it said
    Restoring land to forest or grassland takes up carbon at roughly five to ten tonnes of CO₂ per hectare per year over about a century, with a wide range.
    What it says now
    Restoring land to forest takes up carbon at roughly five to ten tonnes of CO₂ per hectare per year over about a century, with a range of one to twenty, and the IPCC warns that the potential may be overestimated.
    Why it changed
    The IPCC figure is for forest, not grassland, and the same source warns the potential may be overestimated.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L29 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · where we are least sure, the sentence on regional land intensities

    · Correction of an error · published, then changed

    What it said
    a tenfold energy expansion needs between three and sixteen times more area than the global average suggests.
    What it says now
    On the regional intensities that apply where land actually is, a tenfold energy expansion would need from more than three to sixteen times all the land that would be freed, not half of it. Those regional figures come from India and the EU, and the source warns that they hold region by region and must not be applied to the whole world. They also measure something different from the global average: the land a region must take out of crops or forestry, including land displaced elsewhere, not only the land the panels cover.
    Why it changed
    The source's figure, 3.4 to 16 times, compares the regional intensities with the land that would be freed, not with the global average. The same source warns that those intensities hold region by region and must not be applied to the whole world, and that they measure land a region takes out of crops or forestry, not only the land panels cover. The page now says both.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L31 (docs/EVIDENCE-REGISTER.md)
    The energy evidence file, sections C1 and C3 (docs/research/ENERGY-LAND-EVIDENCE-2026-09-19.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentences on what limits the energy transition

    · Correction of an error · published, then changed

    What it said
    Land is not what limits the energy transition. The size of the build is: primary energy is nearly six times today's electricity.
    What it says now
    On the measured footprint, land is not what limits the energy transition. The size of the build is: primary energy is nearly six times today's electricity. Primary energy overstates what an electric system would need, so treat these areas as upper figures.
    Why it changed
    The arithmetic is right, but primary energy is not what an electric system would need, so the figures are upper figures. The claim that land does not limit the transition holds on the measured footprint only: at regional intensities, a larger build needs more land than would be freed. The page now says both.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, rows L28 and L31 (docs/EVIDENCE-REGISTER.md)
    The energy evidence file, section C3 (docs/research/ENERGY-LAND-EVIDENCE-2026-09-19.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · where we are least sure, the sentence on the ceiling on photosynthesis

    · Correction of an error · published, then changed

    What it said
    the literature carries about 4.6% for C3 and 6% for C4 at 30 °C, and about 12% on another basis, and no source we retrieved states how the two relate.
    What it says now
    the literature carries more than one theoretical maximum, each on a different basis, and no source we retrieved states how they relate.
    Why it changed
    The page said it publishes no ceiling and then printed three, without their conditions or sources. The figures are removed.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L34 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on photorespiration

    · Correction of an error · published, then changed

    What it said
    20 to 50% of the yield of C3 crops, with the largest losses in hot, dry places.
    What it says now
    20 to 50% of the efficiency with which C3 crops turn light into biomass, with the largest losses in hot, dry places.
    Why it changed
    The source measures the efficiency of turning light into biomass, not harvested yield.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L08 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · where we are least sure, the sentence on the nitrogen boundary

    · Revision · published, then changed

    What it said
    where the answer we could source puts it at roughly 78% of the application boundary, under rather than over.
    What it says now
    where the answer we could source, a 2020 report from the Potsdam Institute for Climate Impact Research and the Stockholm Resilience Centre, puts it at roughly 78% of the application boundary, under rather than over.
    Why it changed
    The figure was right, and the page now names the report it comes from.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L33 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence after the land chart

    · Correction of an error · published, then changed

    What it said
    That is the largest and most robust of these comparisons, because it is a measurement rather than a model.
    What it says now
    On land per unit of protein, beef from a beef herd needs about seventy-four times the area of tofu, and lamb about eighty-four times.
    Why it changed
    The second sentence is removed. The dataset is a meta-analysis of life-cycle assessments, which are models, and the site's own per-calorie figures give larger ratios. The ratios in the first sentence stand.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L13 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · how this page was made, the line on land figures

    · Correction of an error · published, then changed

    What it said
    Figures for how much land a food needs are global medians, and individual producers vary widely around them.
    What it says now
    Figures for how much land a food needs are global means, and individual producers vary widely around them.
    Why it changed
    The dataset gives global means, not medians. The land chart's caption carried the same error and has its own entry.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L12 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the caption of the chart of land per 100 g of protein

    · Correction of an error · published, then changed

    What it said
    Global medians for 2010; Poore & Nemecek (2018), retrieved as Our World in Data's published dataset.
    What it says now
    Global means, from studies whose median year is 2010; Poore & Nemecek (2018), retrieved as Our World in Data's published dataset.
    Why it changed
    The dataset gives global means, not medians, and 2010 is the median year of the studies behind it, not the year of the figures.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L12 (docs/EVIDENCE-REGISTER.md)
    The corrected caption, in the module that draws the chart (backend/core/why.py)

  • Why food · the sentence on agricultural land

    · Revision · published, then changed

    What it said
    That is more than a third of all land and roughly half of the land that is not ice or desert.
    What it says now
    That is more than a third of all land and almost half of the land that is not ice or desert.
    Why it changed
    The same publisher gives 44% and 50%. "Roughly half" took the upper figure, and "almost half" covers both.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L19 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on protein from grazing cattle

    · Correction of an error · published, then changed

    What it said
    On grazing and forages, cattle need only 0.6 kilograms of protein from human food to produce a kilogram of protein in milk and meat.
    What it says now
    On grazing and forages, the authors of that feed study (Mottet et al., 2017) say in their own summary that cattle need only 0.6 kilograms of protein from human food to produce a kilogram of protein in milk and meat.
    Why it changed
    The figure is held only from the authors' own summary and a press release, not from the paper, so the sentence now says whose statement it is.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L17 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the land, the sentence on grazing land that cannot become cropland

    · Correction of an error · published, then changed

    What it said
    One more thing belongs here, because it limits what "freed land" can mean: much of the land used to graze cattle is grassland that cannot be converted to cropland. Pasture is not cropland waiting to happen.
    What it says now
    One more thing belongs here, because it limits what "freed land" can mean. The authors' own summary of Mottet et al. (2017) says that most of the land used to feed livestock is grassland, with a large share of pastures that could not be converted to cropland and could only be grazed. Pasture is not cropland waiting to happen.
    Why it changed
    The sentence rested on a secondhand quote with no retrieval record, and it credited the paper with a finding this site holds only from the authors' own summary. The summary was read on 22 September 2026, and it says a large share of pastures could not be converted to cropland. The sentence now says whose statement it is and uses its wording.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L37 (docs/EVIDENCE-REGISTER.md)
    The authors' own summary: Mottet and others, More fuel for the food/feed debate, Feedipedia Broadening Horizons No. 47, November 2017, read on 22 September 2026 (https://www.feedipedia.org/content/more-fuel-foodfeed-debate)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentence on crop efficiency in the field

    · Correction of an error · published, then changed

    What it said
    and the best measured field values are about 2.2% for C3 crops and 3.0% for C4.
    What it says now
    A 2026 review by Zhou et al. (J. Integr. Plant Biol. 68:2608) puts the actual efficiency of modern crop canopies at about 2.2% for C3 crops and 3.0% for C4.
    Why it changed
    The sentence read as if the figures were Ort et al.'s. They are from Zhou et al. 2026, which calls them the actual canopy efficiency of modern crops, not the best measured values.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L06 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the sentences on feed conversion

    · Correction of an error · published, then changed

    What it said
    On feed, whole-cycle conversion is about 14.97 pounds of feed per pound of liveweight for beef cattle against 2.19 for broilers. But the feedlot stage alone is 6.35, because most of a beef animal's life is spent eating grass.
    What it says now
    On feed, whole-cycle conversion in US systems is about 14.97 pounds of feed dry matter per pound of liveweight for beef cattle, against 2.19 for broilers. A Canadian benchmark for the feedlot stage alone is 6.35 kilograms of dry matter per kilogram of gain. The two figures differ because the feedlot figure leaves out the grass a beef animal eats for most of its life, and the calf and backgrounding phases before the feedlot.
    Why it changed
    The 6.35 is a Canadian figure in another unit, and it was printed with no unit. The source gives two causes for the gap between the figures, and the page gave one.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L15 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · where we are least sure, feed accounting

    · Correction of an error · published, then changed

    What it said
    Feed accounting. Whether grass and crop residues count as food is a genuine dispute, not a mistake by either side, and it is the single largest reason two honest studies can disagree by a factor of ten on the same animal.
    What it says now
    Removed. The dispute over whether grass and crop residues count as feed is stated in the section on where the numbers are big, and where they are not.
    Why it changed
    No retrieved source gives a factor of ten, or ranks this as the largest reason studies differ.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L36 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · the page's description

    · Correction of an error · published, then changed

    What it said
    Feeding animals to eat them costs several times more land, water and feed than eating plants directly.
    What it says now
    Beef and lamb need fifty to ninety times more land than the plant foods that would replace them. For other animals and other resources the gap is smaller, and on one measure it reverses.
    Why it changed
    "Several times" understated land for beef and lamb and overstated water for milk, eggs and chicken, and on feed that people could eat the comparison can reverse.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L01 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • Why food · where we are least sure, the sentence on capacity factors

    · Revision · published, then changed

    What it said
    Capacity factors inside a single country range from 11.5% to 29.8%.
    What it says now
    Capacity factors inside a single country, the United States, range from 11.5% to 29.8%.
    Why it changed
    The figures were right, and the page now names the country they come from.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L32 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/argument.json)

  • What we know · health comparisons · the sources list, the title of source schwingshackl-2017

    · Correction of an error · published, then changed

    What it said
    Temporal association patterns and dynamics of amyloid-β and tau in Alzheimer's disease. European Journal of Epidemiology, 2018.
    What it says now
    Food groups and risk of all-cause mortality: a systematic review and meta-analysis of prospective studies. The American Journal of Clinical Nutrition, 2017.
    Why it changed
    The source record carried another paper's title. The link, the stored copy and the sentences it backs were always the right paper. The record also named the wrong journal and year; the DOI is the American Journal of Clinical Nutrition, 2017.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, Assessment (c) (docs/EVIDENCE-REGISTER.md)
    The stored copy of the source, which carries its own title (docs/what-we-know/health-comparisons/sources/schwingshackl-2017.txt)
    The corrected source record (docs/what-we-know/health-comparisons/sources.json)

  • What we know · health comparisons · the sources list, the title and kind of source kahwati-2018

    · Correction of an error · published, then changed

    What it said
    Interventions to Prevent Falls in Older Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA. evidence report.
    What it says now
    Vitamin D, Calcium, or Combined Supplementation for the Primary Prevention of Fractures in Community-Dwelling Adults: US Preventive Services Task Force Recommendation Statement. JAMA. recommendation statement.
    Why it changed
    The source record carried another paper's title, and called the paper an evidence report. The link and the stored copy are the US Preventive Services Task Force's recommendation statement, and the sentences it backs quote that statement. The link, the stored copy and the sentences were always the right paper.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, Assessment (c) (docs/EVIDENCE-REGISTER.md)
    The stored copy of the source, which carries its own title (docs/what-we-know/health-comparisons/sources/kahwati-2018.txt)
    The corrected source record (docs/what-we-know/health-comparisons/sources.json)
    The regenerated gate artifact for the entry (docs/what-we-know/health-comparisons/gate-output.json)

  • What we know · health comparisons · the sources list, the title of source hu-1997-nhs

    · Correction of an error · published, then changed

    What it said
    Pathophysiology of premature skin aging induced by ultraviolet light
    What it says now
    Dietary fat intake and the risk of coronary heart disease in women
    Why it changed
    The source record carried another paper's title. The link, the stored copy and the sentences it backs were always the right paper.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, Assessment (c) (docs/EVIDENCE-REGISTER.md)
    The stored copy of the source, which carries its own title (docs/what-we-know/health-comparisons/sources/hu-1997-nhs.txt)
    The corrected source record (docs/what-we-know/health-comparisons/sources.json)

  • What we know · health comparisons · the sources list, the title of source caret-1996

    · Correction of an error · published, then changed

    What it said
    Blood pressure and end-stage renal disease in men
    What it says now
    Effects of a combination of beta carotene and vitamin A on lung cancer and cardiovascular disease
    Why it changed
    The source record carried another paper's title. The link, the stored copy and the sentences it backs were always the right paper.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, Assessment (c) (docs/EVIDENCE-REGISTER.md)
    The stored copy of the source, which carries its own title (docs/what-we-know/health-comparisons/sources/caret-1996.txt)
    The corrected source record (docs/what-we-know/health-comparisons/sources.json)

  • What we know · health comparisons · the short answer, the sentence on what replaces the meat

    · Correction of an error · published, then changed

    What it said
    What replaces the meat matters more than whether the meat is there.
    What it says now
    Our reading is that what replaces the meat matters more than whether the meat is there.
    Why it changed
    The same sentence as in the floor, and the same reason: it is this site's reading of the evidence, not a study result, and it now says so.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H06 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, the sentence on what replaces the meat

    · Correction of an error · published, then changed

    What it said
    What replaces the meat matters more than whether the meat is there.
    What it says now
    We read the evidence as saying that what replaces the meat matters more than whether the meat is there, though in the US study behind it, the links between protein source and death appeared only in people with at least one unhealthy habit, such as smoking, heavy drinking, excess weight or inactivity.
    Why it changed
    No source compares the two directly, so the sentence is this site's reading and now says so. It also carries the study's own limit: the links between protein source and death appeared only in people with at least one unhealthy lifestyle factor.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H06 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, first sentence

    · Correction of an error · published, then changed

    What it said
    You do not need animal products to eat well.
    What it says now
    You do not need animal products to eat well if you are an adult, not pregnant or breastfeeding, and your diet is planned.
    Why it changed
    The sentence named no population and no condition. The source behind it supports a planned diet for adults who are not pregnant or breastfeeding, and no source this entry holds covers pregnancy, infancy or childhood.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H01 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, the sentence on B12 and the nutrients to watch

    · Correction of an error · published, then changed

    What it said
    You do need B12 — from a supplement or from fortified food — and it is worth paying attention to calcium, iodine, iron and omega-3.
    What it says now
    You do need B12, from a supplement or from fortified food. It is also worth paying attention to protein (how much, its quality, and the amino acid lysine), iodine, iron, zinc, calcium, vitamin D, and the omega-3 fats EPA and DHA. The protein part of this entry explains why. Pregnancy, infancy and childhood are outside the sources behind this page.
    Why it changed
    The list named five of the ten nutrients that the entry's own protein section says every adequacy claim must name. It now names all ten.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H03 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, the sentence on nutrition data

    · Revision · published, then changed

    What it said
    We hold that data for none of them, and two of the fifteen state a B12 figure.
    What it says now
    We hold that data for none of them.
    Why it changed
    The count was right for the fifteen products the directory held when it was written. The directory has since grown to 228 products, and nobody has counted B12 figures across them, so the count is removed rather than left stale.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H08 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, the sentence comparing well-built diets

    · Correction of an error · published, then changed

    What it said
    A well-built diet with animal protein and a well-built one without are close to each other.
    What it says now
    In our reading of the large cohorts, a well-built diet with animal protein and a well-built one without are close to each other on heart disease and on death from any cause. They are not close on everything: in a British cohort, vegetarians had 20% more stroke than meat eaters, and vegans more than twice the rate of hip fracture.
    Why it changed
    "Close" named no outcome and hid the two outcomes where the entry's own sources found a difference, stroke and hip fracture. The sentence is also this site's reading, not a single study's result, and now says so.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H04 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • What we know · health comparisons · the floor, the sentence on badly built diets

    · Correction of an error · published, then changed

    What it said
    A badly built version of either is much worse than both.
    What it says now
    Among US health professionals, those who scored highest for less healthy plant foods had 32% more coronary heart disease than those who scored lowest, and those who scored highest for healthy plant foods had 25% less. No study we hold makes that comparison for diets that include meat.
    Why it changed
    "Much worse" had no number, and the support is only for plant-based diets. The sentence now gives the study's result and its population, and says that no source here makes the same comparison for diets that include meat.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row H05 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/what-we-know/health-comparisons/entry.json)

  • How foods compare · the paragraph on protein from grazing cattle

    · Correction of an error · published, then changed

    What it said
    Mottet et al. (2017) found that cattle need only 0.6 kg of protein from human-edible feed to produce a kilogram of protein in milk and meat,
    What it says now
    The authors' own summary of Mottet et al. (2017) says that cattle need only 0.6 kg of protein from human-edible feed to produce a kilogram of protein in milk and meat,
    Why it changed
    "Found" credited the paper with a figure this site holds only from the authors' own summary and a press release.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row C07 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/comparison.json)

  • How foods compare · the point that land is not one thing

    · Correction of an error · published, then changed

    What it said
    Much of the land cattle graze is grassland that cannot be converted to cropland, so it could not simply be turned into a field of peas. Mottet et al. (2017) report that most of the land livestock use is grassland, much of which can only be grazed.
    What it says now
    The authors' own summary of Mottet et al. (2017) says that most of the land used to feed livestock is grassland, with a large share of pastures that could not be converted to cropland and could only be grazed. That land could not simply be turned into a field of peas.
    Why it changed
    The sentence rested on a secondhand quote with no retrieval record, and it credited the paper with a finding this site holds only from the authors' own summary. The summary was read on 22 September 2026, and it says a large share of pastures could not be converted to cropland. The sentence now says whose statement it is and uses its wording.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row L37 (docs/EVIDENCE-REGISTER.md)
    The authors' own summary: Mottet and others, More fuel for the food/feed debate, Feedipedia Broadening Horizons No. 47, November 2017, read on 22 September 2026 (https://www.feedipedia.org/content/more-fuel-foodfeed-debate)
    The corrected text, in the file the page renders (docs/why/comparison.json)

  • How foods compare · the note on which foods each chart covers

    · Correction of an error · published, then changed

    What it said
    The three datasets do not cover the same foods. A food absent from a chart is absent from that dataset, not zero in it.
    What it says now
    The three datasets do not cover the same foods, and the land and water charts each show a selection of their dataset's rows rather than all of them. A food absent from a chart is not zero, and it may still be in the dataset.
    Why it changed
    The land and water charts are selections from larger datasets. The calorie dataset holds dairy-herd beef and other pulses, which the chart leaves out. The feed chart shows every row of its dataset.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The evidence register of 22 September 2026, row C02 (docs/EVIDENCE-REGISTER.md)
    The corrected text, in the file the page renders (docs/why/comparison.json)

  • Why food · the sentence on what the world's energy supply would take in land

    · Correction of an error · changed before it was published

    What it said
    roughly half a percent of the land that would be freed, and around one percent of the world's cropland.
    What it says now
    roughly five percent of the land that would be freed, and around nine percent of the world's cropland.
    Why it changed
    Both shares were about ten times too small, and the page refuted itself two paragraphs later. The area was right; the percentages came from three table rows in the energy evidence file that were a factor of ten too small, while the same file's ten-times and hundred-times rows were correct. The audit's correction found three further instances of the same slip in that evidence file.
    Served before the change
    No. This sentence was never served to a reader in the form it is quoted here.
    Evidence
    The research audit of 21 September 2026, finding E1 (docs/research/RESEARCH-AUDIT-2026-09-21.md)
    The energy evidence file's own dated correction block (docs/research/ENERGY-LAND-EVIDENCE-2026-09-19.md)
    The corrected sentence, in the argument file the page renders (docs/why/argument.json)

  • How foods compare · four figures in the prose and their source records

    · Revision · changed before it was published

    What it said
    Four figures — 0.6 kg of human-edible protein, 86% of livestock dry matter, soymeal at 4% of what livestock eat, and about 65% of cattle-grazed land being unsuitable for crops — rested on two source records with no retrieval recorded against them, on a page that said every figure was retrieved on 19 September 2026.
    What it says now
    The first three figures are recorded against the paper's abstract and the authors' own summary, both read on 21 September 2026, and the page's own note says so. The 65% figure is withdrawn rather than restated, because the page it was attributed to does not carry it.
    Why it changed
    This is a revision rather than an error in the figures: the 0.6 kg, 86% and 4% figures are correct against the published record, and better information made them verifiable. The 65% is the part that had to come out, because its attribution could not be confirmed and the paper's own tables were never read.
    Served before the change
    No. This sentence was never served to a reader in the form it is quoted here.
    Evidence
    The research audit of 21 September 2026, finding E3 (docs/research/RESEARCH-AUDIT-2026-09-21.md)
    The source record and the withdrawn figure, in the comparison file (docs/why/comparison.json)

  • How foods compare · the paragraph conceding what grazing systems contribute

    · Correction of an error · changed before it was published

    What it said
    Mottet et al. (2017) estimated that grazing systems return about 0.6 kg of human-edible protein per kg of milk and meat protein — roughly 1.7 to 1.
    What it says now
    Mottet et al. (2017) found that cattle need only 0.6 kg of protein from human-edible feed to produce a kilogram of protein in milk and meat, so a grazing system can add to the food supply rather than only subtract from it.
    Why it changed
    The ratio was inverted. Mottet's finding is that cattle consume 0.6 kg of human-edible protein to produce 1 kg of protein, and the old sentence's own arithmetic did not work: a system returning 0.6 per 1 is 0.6 to 1, not 1.7 to 1. The site's Why food page already carried the figure the right way round, so two of its pages stated one number in opposite directions. The paragraph that gives the other side its strongest case is the last place this should have gone wrong.
    Served before the change
    No. This sentence was never served to a reader in the form it is quoted here.
    Evidence
    The research audit of 21 September 2026, finding E2 (docs/research/RESEARCH-AUDIT-2026-09-21.md)
    The corrected paragraph, in the comparison file the page renders (docs/why/comparison.json)

  • The animal's case · the reference-diet table, in a draft that has not been served

    · Correction of an error · changed before it was published

    What it said
    A table of the EAT-Lancet Commission's reference diet, using the 2019 edition's figures and not naming which edition it was.
    What it says now
    The same table, labelled as the 2019 edition's Table 1, with the later revision's rows named beside it: red meat as one row at 15 g a day, and legumes at 75 g a day.
    Why it changed
    Both editions are accurate for their own edition, but this site's health entry quotes the later revision. Two of its own pages would have shown different numbers for the same diet without saying why, which reads as a contradiction rather than as an edition. The fix was made before the page was ever served and before it was published anywhere.
    Served before the change
    No. This sentence was never served to a reader in the form it is quoted here.
    Evidence
    The research audit of 21 September 2026, the ethics section and the corrections applied at revision 2 (docs/research/RESEARCH-AUDIT-2026-09-21.md)
    The corrected draft (docs/research/MORAL-CASE-ARGUMENT.md)

  • The animal's case · the affordability paragraph, in a draft that has not been served

    · Correction of an error · changed before it was published

    What it said
    dried lentils cost 1.64 cents per gram of protein against ground beef at 9.60 cents — a little under one sixth.
    What it says now
    dried lentils cost 1.64 cents per gram of protein against ground beef at 9.60 cents — about a sixth.
    Why it changed
    1.64 divided by 9.60 is 17.1%, which is a little over one sixth, and the same draft says "about a sixth" in two other places. The direction of the error flattered the argument slightly, which is the kind this page exists to catch. The fix was made before the page was ever served.
    Served before the change
    No. This sentence was never served to a reader in the form it is quoted here.
    Evidence
    The research audit of 21 September 2026, the ethics section and the corrections applied at revision 2 (docs/research/RESEARCH-AUDIT-2026-09-21.md)
    The corrected draft (docs/research/MORAL-CASE-ARGUMENT.md)

  • What we know · the tenth limit on all four entries: cultivated meat, precision-fermented protein, mycoprotein and biomass, and algae and seaweed

    · Correction of an error · published, then changed

    What it said
    No person has approved a single claim, which is why the page is not released.
    What it says now
    No person reviewed or approved these sentences before publication.
    Why it changed
    The second half of the sentence was false. All four entries were released, served and indexable from 20 September 2026, and the first half is still true: no person has decided any of their claims. The limit now says what is the case — the pages were published without per-claim human approval.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The research audit of 21 September 2026, section 5, and the four entries' own limits blocks (docs/research/RESEARCH-AUDIT-2026-09-21.md)

  • What we know · cultivated meat · the short answer on the entry's first page

    · Correction of an error · published, then changed

    What it said
    Regulators in two countries have cleared it, and a person outside Singapore and the United States will struggle to find it.
    What it says now
    Regulators in at least four countries have cleared it, and a person outside those markets will struggle to find it.
    Why it changed
    The count was wrong: Israel approved cultivated beef in January 2024, and Australia and New Zealand cleared cultivated quail in 2025. The entry's own sentences already said three countries, so the page contradicted itself one section below its own short answer.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The research audit of 21 September 2026, finding T1 (docs/research/RESEARCH-AUDIT-2026-09-21.md)

  • What we know · cultivated meat · the sentence about Vow's staff and leadership

    · Correction of an error · published, then changed

    What it said
    Vow cut 30% of its staff and changed chief executive in June 2026.
    What it says now
    Vow changed chief executive in June 2026 and cut jobs, after laying off 30% of its staff in 2025, while keeping its products on sale.
    Why it changed
    The sentence moved a 2025 layoff into 2026. The report it cites says the June 2026 change followed layoffs from 2025 that affected 30% of staff, and it does not put a number on the later job cuts.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The research audit of 21 September 2026, finding T3, and the regenerated gate artifact for the entry (docs/what-we-know/cultivated-meat-and-seafood/gate-output.json)

  • What we know · cultivated meat · the sentence about the FDA's animal-cell inventory

    · Correction of an error · published, then changed

    What it said
    Those five cover cultured chicken cell material from two sponsors, cultured pork fat cells from one and cultured salmon cell material from one.
    What it says now
    Those five cover cultured chicken cell material from three sponsors, cultured pork fat cells from one and cultured salmon cell material from one, with FDA response dates from November 2022 to July 2025.
    Why it changed
    The sentence counted the list it quoted. The retrieved inventory entry names three chicken sponsors — Believer Meats, UPSIDE Foods and GOOD Meat — and two plus one plus one did not add up to the five the sentence claimed.
    Served before the change
    Yes. This sentence was served to readers before it was changed.
    Evidence
    The research audit of 21 September 2026, finding T2, and the regenerated gate artifact for the entry (docs/what-we-know/cultivated-meat-and-seafood/gate-output.json)

Tell us about one we have not made

If a source we used has changed, moved, been corrected or been retracted, tell us and we will check the sentence that rests on it and record what we find here. A sentence we cannot stand behind comes out of the page and its withdrawal is recorded on this page rather than being quietly deleted.

Every date on this page is rendered the same way — day, month name, year — 21 September 2026. The stored dates behind them, and the machine-readable dates, are unchanged.

Request a correction ↗ · What we know → · What a listing does not tell you