Corrections

Minor text corrections: PRISMA record counts updated (141→152 records, 12→13 duplicates, 120→126 excluded), abbreviation MeSH→MESH, deleted redundant phrase, Cochrane CENTRAL→CDSR. No data or conclusion changes. Source: BMC Complementary Medicine and Therapies 2021

The journal's own notice, recorded on the study page

Article has been corrected (BMJ 2020 Nov 3;371:m4264). Correction related to disclosure of author conflicts of interest. Source: The BMJ 2018

The journal's own notice, recorded on the study page

A corrigendum (DOI: 10.1016/j.exger.2020.110897) corrected incorrect attribution to Psilander et al. (2019) regarding myonuclear loss during detraining, clarifying that this finding came from a secondary analysis by Murach et al. (2019), not the original Psilander study. The corrigendum also notes that Eftestøl et al. (2020) raised concerns about the Murach reanalysis, and updates the discussion interpretation accordingly. Additionally corrects Laurentani et al. year from 2020 to 2003 and co-author affiliations. Source: Experimental Gerontology 2020

The journal's own notice, recorded on the study page

The study page and its audio said five days in the fridge cost fresh green beans 40% of their vitamin C while the frozen bag 'hadn't shifted'. The claim said the broccoli in the fridge 'was losing' its nutrients, that fridge-stored corn 'lost over half its beta-carotene', that frozen produce gave the same nutrient levels whatever the season or store, and that vitamin C was perfectly consistent across both studies. The Short's audio said spinach loses 75% of its vitamin C by Wednesday, three days after it is bought.

The 40% is the gap between fridge-stored and frozen green beans: frozen beans had more vitamin C even than fresh beans bought the same day, and five days in the fridge cost 13%, a drop that was not statistically significant. The pages now say that most comparisons showed no significant difference, that frozen came out ahead more often than the reverse, and that fridge-stored broccoli kept more beta-carotene and folate than frozen, and fridge-stored spinach more vitamin C. The Short's audio gives the 75% as the loss over a week.

Li et al. (2017), Tables 1 to 3: green beans had 7.7, 6.7 and 11.0 mg of vitamin C per 100 g fresh, fridge-stored and frozen; fridge-stored broccoli had 1,710 µg of beta-carotene per 100 g against 940 frozen; spinach had less vitamin C frozen than fresh or fridge-stored. The study pooled produce from six supermarkets and reports no season-by-season or store-by-store comparison. The 75% spinach figure is a seven-day loss at 4 °C (Favell, 1998, as reported by Rickman et al., 2007). Our pages turned comparisons into losses and left out the results that went the other way. Source: Li L, Pegg RB, Eitenmiller RR, Chun J-Y, Kerrihard AL. Selected nutrient analyses of fresh, fresh-stored, and frozen fruits and vegetables. J Food Compos Anal 2017;59:8-17.

Live check of the 25 September 2026 correction on the freezing Short: every page citing the same papers was checked against the papers themselves, 2026-09-25

The study page called the University of Georgia team behind Li et al. (2017) 'an independent team' and its result 'independent confirmation', its audio said the research came from 'independent teams', and it said the UC Davis papers stand on 'independent academic funding'. The claim described 'two independent laboratory analyses' and 'two independent university laboratories'.

Both pages now say 'two separate' teams or analyses and state that the Georgia study was funded by the Frozen Food Foundation, which also chose the fruits and vegetables it tested. The study page now says the UC Davis papers name no funder and declare no competing financial interest, and that their seeds were donated by Seminis Vegetable Seed Co.

Li et al. (2017) state: 'This work was supported by a grant from the Frozen Food Foundation', and that the choice of produce 'was predetermined by the Frozen Food Foundation (FFF), the funding agency for this study'. Both pages disclosed that funding elsewhere, but 'independent' beside an industry-funded study tells a reader the opposite, and our rule is that industry-funded research is never called independent. The two Bouzari et al. (2015) papers name no funder at all, so 'independent academic funding' was a claim they do not make. Source: Li L, Pegg RB, Eitenmiller RR, Chun J-Y, Kerrihard AL. Selected nutrient analyses of fresh, fresh-stored, and frozen fruits and vegetables. J Food Compos Anal 2017;59:8-17.

Live check of the 25 September 2026 correction on the freezing Short: every page citing the same papers was checked against the papers themselves, 2026-09-25

The Short's summary and answer said there were 'two exceptions' (beta-carotene and myrosinase), credited both to a review that covers neither, and said everything that survives blanching holds in the freezer. The study page said the carrots' fiber loss was not caused by frozen storage, named iron as the only mineral that dropped, gave leaching as the cause of the iron drop, and said the UC Davis team thinks the heat 'cracked open cell walls'. The claim called beta-carotene 'the single clear exception', said blanching causes 'minor' iron leaching and that nutrients are 'locked in at processing', and said vitamin E, calcium and copper show no loss during frozen storage.

All three pages now name the exceptions the papers report: beta-carotene fell by more than half over 90 days in frozen peas, spinach and carrots; frozen spinach and one of two carrot varieties had 10 to 45% less iron than fresh; frozen peas, spinach and corn had about 7 to 18% less magnesium; and both carrot varieties lost about a quarter of their fiber in frozen storage. The iron drop is placed at the blanching step, with the authors' two possible causes, iron leaching out or soil washing off. The rise in vitamin E is described as the authors describe it, as a possible effect of steam blanching. Summaries now say 'most nutrients' and 'the biggest exception', and the Short's summary box credits each source it draws on.

Checked line by line against the two companion papers by Bouzari, Holstege and Barrett (2015) and, for myrosinase, Dosz and Jeffery (2013). It is the same mistake as our correction of the Short's minerals FAQ earlier on 25 September 2026 (2026-09-25-frozen-vegetables-some-minerals-and-fiber-did): lines written from the abstract turned 'most showed no difference' into 'nothing changed'. That correction fixed only the sentence it was found in. When it went live, we checked every page citing these papers against the papers themselves, and found these lines. The study also shows vitamin E slipping during frozen storage in broccoli and carrots, and it measured calcium and copper only right after processing, so 'no loss during frozen storage' was more than it showed. Source: Bouzari A, Holstege D, Barrett DM. Mineral, Fiber, and Total Phenolic Retention in Eight Fruits and Vegetables: A Comparison of Refrigerated and Frozen Storage. J Agric Food Chem 2015;63:951-956.

Live check of the 25 September 2026 correction on the freezing Short: every page citing the same papers was checked against the papers themselves, 2026-09-25

Pages about Guarneiri et al. 2024 said it appeared in Nutrition Reviews and that digestion is about 3% of daily energy expenditure. They presented the review's 72 kcal a day, one trial's difference in total daily energy expenditure (about 3%), as the extra calories it takes to digest a high-protein meal or as 'protein's thermic advantage', sometimes 'at typical intakes', 'the energy cost of a 30-minute walk' or a '72-120 kcal' range. They said the advantage 'disappears', 'vanishes' or is 'gone' during a calorie deficit or on longer diets, where the paper found it was not statistically significant, and in other places said a meal's extra heat was still produced on long-term diets, which the paper did not find. They called it 'the largest meta-analysis ever conducted' and protein's effect 'the largest of any food'. They credited the commonly cited 20-30% digestion cost, and our calorie arithmetic built on it, to the 52 studies, and said the effect held or was robust 'across study designs and populations'. Three citations carried invented titles and one an invented co-author, a claim page quoted the paper with sentences that are not in it, the study page called the first author's firm independent, and it credited another study's insulin finding to this paper.

The journal is Advances in Nutrition and digestion's share is about 10%, as the paper says. The 72 kcal is described as what it is: in one trial the review uses as an example, a higher-protein diet raised total daily burn by 72 calories (about 3%), and on longer diets the review credits such rises mainly to resting metabolism, not digestion. Results that were not statistically significant are called that, with their numbers (for example, 4 calorie-deficit comparisons, P = 0.407). The paper is a 2024 meta-analysis of 52 studies, which its authors describe as the first to cover both single meals and longer-term diets. The 20-30% range is labelled as the commonly cited figure and our arithmetic as ours. Citations carry the paper's real title and its six authors, the claim quotes the paper's own conclusion, and the study page says the first author's firm reports funding or consulting fees from General Mills and other food companies.

The paper cited was always the right one; the descriptions around it were written from our notes and earlier drafts instead of being checked line by line against the paper. A scan while fuelling a new Short on whole-food digestion found the wrong journal, the 3% and the 72-calorie misreading on 5 pages; reading the full paper for the fix found the rest. Four pages' audio repeated some of these statements; it was regenerated with the fix. Source: Guarneiri LL, Adams CG, Garcia-Jackson B, Koecher K, Wilcox ML, Maki KC (2024). Effects of Varying Protein Amounts and Types on Diet-Induced Thermogenesis: A Systematic Review and Meta-Analysis. Advances in Nutrition 15(12):100332

SF1 fuelling of the Short 'do-you-burn-more-calories-digesting-whole-foods' (KW-979), then a full read of the paper for the site-wide fix, 2026-09-25

The FAQ 'Do frozen vegetables lose minerals and fiber?' ended: 'Minerals and fiber hold steady either way.'

It now names the drops its source reports: frozen spinach and one carrot variety had 10 to 45% less iron than fresh, frozen peas, spinach and corn about 7 to 18% less magnesium, and both carrot varieties lost about a quarter of their fiber in frozen storage.

The FAQ's source, Bouzari, Holstege and Barrett (2015), found no significant fresh-versus-frozen difference for most of the eight fruits and vegetables it tested, and that part of our answer was right. The same paper reports these exceptions: 'in spinach and carrot variety A, the frozen samples had between 10% and 45% less iron than the fresh samples'; 'Frozen peas, spinach, and corn contained between ~7% and 18% less magnesium'; 'Both varieties of carrots lost fiber (~25%) during frozen storage'. Our answer turned 'most showed no difference' into 'hold steady either way'. It was written from the paper's abstract, which says results depended on the commodity but does not list the exceptions, and our checks did not test the sentence against the full paper. The error was found when a new Short on raw versus cooked spinach, which reports the same paper's spinach result, was checked against the rest of the site. Source: Bouzari A, Holstege D, Barrett DM. Mineral, Fiber, and Total Phenolic Retention in Eight Fruits and Vegetables: A Comparison of Refrigerated and Frozen Storage. J Agric Food Chem 2015;63:951-956.

SV1 cross-content check (Gate 1C) for the Short is-raw-spinach-better-than-cooked-for-iron (KW-975), 2026-09-25

The page said that whole-food meals needed nearly twice the digestive energy of processed-food meals 'made with the same macronutrients' (narrative) and 'made from the same ingredients' (Weight of Evidence card). The card added that food form may be the bigger factor in how much nutrition the body extracts.

Both places now say that a whole-food sandwich meal cost nearly twice the digestive energy of a processed one with the same calories, and that the whole-food meal also carried more protein and about three times the fibre. The card adds that this was one meal pair and that the authors urge caution in extending it to other meals.

Barr and Wright (Food & Nutrition Research, 2010) matched their two cheese-sandwich meals for calories only. The whole-food meal was 20% protein and 40% carbohydrate against 15% and 50% for the processed one, with about three times the fibre, and the breads and cheeses differed by design. The paper measured the energy spent after eating, not how much nutrition the body extracts, and its authors ask for caution in extending a single meal comparison to other meals. Our error: the study's research brief described the meals as matched for macronutrients, and the page repeated the brief instead of the paper. Source: Barr SB, Wright JC. Postprandial energy expenditure in whole-food and processed-food meals: implications for daily energy expenditure. Food Nutr Res. 2010;54:5144.

FitChef's Short Fueler, reading the paper while preparing a new Short on the same question (2026-09-25)

The guide's headline statistic was corrected to 8,666 participants on 2026-06-03 (see the record of that date), but nine other fields kept the old figure of 9,350: the opening hook, the key takeaway, two body paragraphs, the skeptic note, a citation capsule, the AI citation hint, the search description and a visual's description.

Every field says 8,666, the sum of the verified participant counts of the six studies in the guide that report one. The seventh, Schoenfeld 2017, reports no total, which is what the 2026-06-03 record explains. The audio script never carried the figure.

The 2026-06-03 fix changed the hero statistic and the cross-references but did not sweep the guide's prose and metadata for the same figure. Found on 2026-09-24 when every correction record was checked against its live page, the day the records went public.

Check of every correction record against its live page, 2026-09-24

The page said the 2025 meta-analysis found that trained lifters need more weekly sets than beginners for growth, presented the strength difference (3 to 9 sets of the lift for well-trained powerlifters against about 1 set) as settled, explained it with a mechanism the paper does not contain, said most studies measured muscle thickness, and called the analysis the largest ever published.

For muscle growth, the analysis found no consistent difference between new and trained lifters in how much each extra weekly set added, and its authors call every such moderator result hypothesis-generating. For strength, the evidence is thinner: one cited powerlifter study found 3 to 9 direct sets of the lift were needed where 1 to 3 were not, and the authors say it warrants further investigation. The invented mechanism, the muscle-thickness claim and the superlative are gone, and the FAQ, summary, speakable, evidence pairs and descriptions were updated to match.

Every line was checked against the paper and its OSF supplement (the training-status plots). Our own source notes and verdict had already overstated the strength split, and the verification gate passed a page whose mechanism paragraph contradicted its sources. This is the same page whose citation had been invented (see the record on the Pelland 2025 citations). Source: Pelland, Remmert, Robinson, Hinson and Zourdos (2025). The Resistance Training Dose-Response: Meta-Regressions Exploring the Effects of Weekly Volume and Frequency on Muscle Hypertrophy and Strength Gain. Sports Medicine 56(2):481-505

Visual Team grounding check (VT step S7), 2026-09-24

Citations of Pelland et al. 2025 (Sports Medicine) carried a made-up paper title on 29 pages, and on some of them made-up co-authors or first names. On one page the citation also gave a wrong journal volume, described the method as a Bayesian network meta-analysis, and listed 16 authors where the paper has 5. Twelve pages called it 'the largest volume analysis ever published', which neither the paper nor its abstract says.

Every citation carries the paper's real title, 'The Resistance Training Dose-Response: Meta-Regressions Exploring the Effects of Weekly Volume and Frequency on Muscle Hypertrophy and Strength Gain', and, where the author list had been invented, its five real authors (Pelland, Remmert, Robinson, Hinson and Zourdos). The superlative now reads 'a 2025 analysis', or the page's own study count (67 studies, 2,058 participants).

The paper cited was always the right one; the words around it were composed rather than copied from the grounded record. Two of our agents write citation titles into page schema without a step that copies them from the locked source identity. A first scan matched only citations carrying the DOI, which is why 12 pages that cite the paper through our own study page were found a day later; the site-wide identifier gate on our roadmap must match those too. Found and fixed 2026-09-24. Source: Pelland, Remmert, Robinson, Hinson and Zourdos (2025). The Resistance Training Dose-Response: Meta-Regressions Exploring the Effects of Weekly Volume and Frequency on Muscle Hypertrophy and Strength Gain. Sports Medicine 56(2):481-505

Visual Team grounding check on the Short 'experienced-lifters-need-more-sets' (VT step S7), then a site-wide scan, 2026-09-24

The page said browning 'barely moves below about 20°C', that powder stored cool 'kept its nutritional profile well beyond eighteen months', that warm storage cost 'a meaningful share of its most important amino acids in under a year', that the chunks were 'browning products binding protein strands together', that there was 'no mold', and that nobody had checked whey isolate. Three of these also sat in the summary, speakable and research fields that search engines and AI assistants read.

The coolest room the study tested was about 21°C, and it sets no threshold. Shelf life was judged by appearance: 9 months at most at 35°C, and at least 18 months in the cooler rooms, which is when the study stopped. Lysine, the one amino acid measured, fell over a year of storage in the higher-lactose powder, and the paper does not report it by temperature. Chunks can form when milk sugar crystallizes as it takes up moisture. Mold and bacteria stayed at low counts in sealed bags. The paper itself cites a storage study of whey isolate.

Before a data visual was added, the page was checked line by line against its only source, a USDA storage study of whey protein concentrate (Tunick et al. 2016). Six statements went beyond the paper. In our process, the writing step introduced a threshold that was in neither the source notes nor the paper, and nothing on file showed that the fact-check gate had run on this page. That gate now has to prove itself on file for every Short (the linked system change). Source: Tunick MH et al. (2016). Journal of Dairy Science 99(3):2372

Visual Team grounding check (VT step S7), 2026-09-24

Four of the DOIs in the evidence citations on the alcohol dose-threshold claim were wrong: two pointed at real but unrelated papers and two did not exist. One PubMed link and two citation titles were also wrong, including inside the page's machine-readable markup.

Every DOI, link and title on the page is verified against the Crossref registry. The other 17 DOIs in the cluster were checked and found correct, with exact title matches.

The citation metadata had been reconstructed from memory instead of copied from verified source files during a fragmented build. The findings and numbers on the page were never affected. A DOI-verification gate is designed and queued for the pipeline.

Internal cross-pipeline checks (the trust audit, then the library supporting-fields agent), resolved by a dedicated DOI audit, 2026-08-09

The Wolf 2023 partial-versus-full range-of-motion meta-analysis showed 24 participants (its study count) on its study page and in five evidence entries of the claim that cites it. The Rosa 2023 single-versus-multi-joint meta-analysis showed 7 in four evidence entries of its claim.

Wolf 2023 shows 658 participants (the sum of the 23 individual study sizes its Table 1 reports) and Rosa 2023 shows 193.

When a meta-analysis does not state a pooled total, the extraction step recorded the number of included studies instead, and the value passed through synthesis and both import builders unchecked. A participant-count check in the library pipeline caught it. The extraction, verification and import specs were changed so it cannot recur.

Library pipeline kill switch KS-LP1-04, 2026-07-27

Two claim pages and the supplements guide attributed the 2024 creatine body-composition meta-analysis (143 randomised trials) to 'Bonilla et al.'.

All 36 references name the paper's actual first author, Pashayee-Khamene et al. The study page itself was already correct.

During cluster planning, a prominently cited reference author (Bonilla appears as reference 1 in the paper) was confused with the paper's authors. Verified against PubMed (PMID 39042054), the publisher's page and the full text. Source: Pashayee-Khamene F, Heidari Z, Asbaghi O, et al. (2024)

Mark's manual review questioned the authorship, 2026-06-27

The Short on whether muscle turns to fat called muscle 'the strongest predictor of calorie burn'. Its source says fat-free mass, which includes organs and bone, not muscle alone, and a sibling page correctly notes that muscle is metabolically quiet at rest (about 6 calories per pound per day). Readers could see two FitChef pages saying opposite things.

The page says fat-free mass is the strongest anthropometric correlate of total energy expenditure, matching its source and the sibling page.

Both pages were grounded in real studies; the wording on one drifted from the source's own term. A cross-content consistency gate now checks every new page against live content for contradictions (the linked system entry).

Manual audit, 2026-06-05

The guide's headline participant total was taken from a single claim's evidence set, most of it one 68,128-participant study, rather than from the studies the guide itself analyses.

The headline total is computed from the guide's own study set, and a verification step blocks any headline stat that cannot be reconstructed from the studies shown on the page.

Internal audit found the borrowed figure made the headline unrepresentative of the guide as a whole.

Internal audit, 2026-06-03

The stuffed portobello mushrooms recipe displayed a 'High Protein' badge. Protein provides 16.7% of its energy, below the 20% that the EU nutrition-claim regulation (1924/2006) requires for 'high protein'.

The badge reads '35g Protein', a factual statement rather than a regulated claim. The recipe pipeline now checks the threshold before assigning any nutrition badge.

Found by the deep pre-scale audit's EU health-claim compliance check.

Deep pre-scale audit, check 17 (EFSA health-claim compliance), 2026-06-03

The Schoenfeld 2017 light-versus-heavy-load meta-analysis page stated '684 participants' across its 21 studies, in the hero, the introduction, the search description and the citation fields. The training guide's headline participant total (9,350) and a cross-reference in the Zhang 2025 exercise-and-deficit study used the same figure.

The participant count is removed and the page keeps '21 studies', which is what the paper reports. The training guide's headline total is 8,666, and the Zhang 2025 cross-reference no longer carries the figure.

Checked against the full paper (Journal of Strength and Conditioning Research 31(12):3508-3523): it reports no pooled participant total. Its Table 1 sums to 630, but that includes control groups excluded from the analysis, so the analysed number is lower and unreported; 684 appears nowhere in the paper. An unverifiable number is treated the same as a fabricated one and removed rather than replaced with an estimate. The figure had been written as a clean number in the correction of 2026-05-25. Source: Schoenfeld BJ et al. (2017). Journal of Strength and Conditioning Research 31(12):3508-3523

Deep pre-scale audit, check 1 (study number traceability), which verified 17,599 numbers across 224 pages, 2026-06-03

The sugar-and-body-weight meta-analysis page, and the claim that cites it, showed 'Participants: 68'. Sixty-eight is the number of studies the meta-analysis pooled (30 randomised trials and 38 prospective cohort studies), not a participant count.

The figure is removed from both pages. The paper reports per-analysis effect sizes rather than one participant total, so no number replaces it.

Internal audit traced the value to the study count. Rather than invent a total, we removed it. Source: Te Morenga et al. (2013). BMJ

Deep pre-scale audit, 2026-06-03

The headline on the Hall 2019 ultra-processed-food study read '340 extra calories'. The 340 figure is how many calories the measured eating rate delivers in a 20-minute meal; 'extra' implied it was the difference between the two diets, which is 508 calories a day.

The headline reads '340 calories', without 'extra'. The 508-calorie daily difference stays in the key finding, where it was already stated correctly.

Internal audit found the word turned a grounded number into a misleading one.

Internal audit, 2026-06-03

On the night-time carbs page, a finding from Sofer et al. 2011 (28% more weight loss with carbohydrates at dinner) was attributed to Gardner et al. 2018. On the processed-food eating-speed page, a Forde finding was attributed to Hall.

Each pull-quote names the study it describes. A field now maps every pull-quote to its source, and a verification check confirms the match before a Short is published.

The template attributed the pull-quote to the first study in the page's source list, whichever study the sentence described. Found by the Shorts scale audit.

Shorts scale audit, 2026-06-02

The evidence sidebar on three guides showed participant counts that did not come from the studies' verified extraction data. In the supplements guide: 2,400 for a caffeine meta-analysis of 149 people, 814 for a BCAA review of 511, 3,600 for a creatine meta-analysis of 3,655, and 4,937 for a study of 4,755. In the protein guide: 24 for a study of 108. In the carbs guide: 1,779 for a glycaemic-index meta-analysis of 2,344.

All six sidebar counts match the verified extraction data for each study, and each guide's sidebar total equals the total in its article text.

The agent that builds the sidebar had no mandated source for participant counts, so it filled them in without one. The article text on every guide was checked and found correct; the error was confined to the sidebar layer. A rule now requires every sidebar count to come from the same grounded extraction files that feed the article (the linked system entry).

An independent AI cross-check flagged the mismatch between article text and sidebar; internal audit, 2026-05-25

On seven study pages the participant count was stored as free text (for example '49 studies, 1863 participants' or 'n=20'), and the page displayed the first number it found, so a reader could see a study count, or zero, where the participant count belongs. On three claim pages the total-participant figure was wrong.

Each study page shows its verified participant count (for example 1,863 for the Morton 2018 protein meta-analysis, 3,877 for the viscous-fibre meta-analysis, 2,359 for the fat-burner meta-analysis, 20 for the fasted-cardio trial and 511 for the BCAA review). The three claims show 116, 303 and 11,113 participants, each checked against the studies they cite. One of the seven, the light-versus-heavy-load meta-analysis, was set to 684 here and later found to be unverifiable; see the record of 2026-06-03.

A number-consistency audit found the rendering system silently misreading free-text sample sizes. Each value was re-checked against the study's grounded extraction before being written as a clean number.

Number consistency audit, 2026-05-25