Food logging accuracy in fight camp

The size of the error, and who measured it

Doubly labelled water is the reference standard here, and it matters why. A subject drinks water enriched with stable isotopes of hydrogen and oxygen; the differential rate at which those isotopes disappear from the body gives carbon dioxide production and therefore total energy expenditure, over one to two weeks of ordinary free living. In an athlete whose body mass is stable, or whose change in body mass is accounted for, expenditure gives energy intake. Nobody has to be believed.

Capling and colleagues gathered 18 validation studies in athletes, 683 participants, 55% male, mean age 21.8 ± 2.6 years. Eleven of those studies used doubly labelled water. Every one of them reported a lower mean energy intake from self-report than from the isotope method: 19% lower on average, with a spread from 0.4% to 36% across studies, and a weighted mean difference of −2793 ± 1134 kJ/day. The meta-analysis found a large pooled effect size of −1.006 (95% CI −1.3 to −0.7).

Two things about that spread deserve more attention than the headline. First, 0.4% and 36% are not the same finding. An athlete at the low end has a log that is close to correct; an athlete at the high end is missing more than a third of what they eat. Applying the mean to an individual is exactly the error this article is about. Second, the sport breakdown — six mixed-sport studies (n = 461), nine aerobic or endurance studies (n = 184), two team-sport studies (n = 26), one aesthetic/ballet study (n = 12) — contains no boxing, no MMA, no muay thai, no BJJ, no wrestling and no judo.

In adults generally, the picture from Burrows and colleagues is the same shape with more detail. Across 59 studies and 6298 adults, with doubly labelled water as the reference in every case, under-reporting by method ran: food records 11–41%, 24-hour recalls 8–30%, diet history 1.3–47%, food-frequency questionnaires 4.6–42%. Food records were the commonest method — 36 studies, twelve of them using weighed records — and 24-hour recall showed both the lowest total amount and the lowest variation of under-reporting, which is not the result most people expect from a method that relies on memory.

The same review found adults with overweight or obesity under-reported more than normal-weight participants, consistently across recall, diet-history and food-record methods. On sex, it found something less quotable and more honest: women showed a greater tendency than men to misreport on 24-hour and multiple-pass recalls, but for food-frequency questionnaires the direction was inconsistent — three studies found men worse, two found women worse — and for food records it was inconsistent too. There is no clean finding that women under-report more. The sex differences in the literature are method-dependent and inconsistent, and printing a flat version of them would be wrong.

Both halves of the equation fail in the same direction

The 1992 New England Journal of Medicine paper by Lichtman and colleagues is the study everyone half-remembers, and it is usually cited wrongly. It is not evidence about people in general.

The subjects were ten treatment-seeking adults with obesity — nine women and one man — selected out of 224 consecutive patients specifically because they described themselves as diet-resistant, and studied for fourteen days using indirect calorimetry and body-composition measurement. They under-reported actual food intake by an average of 47 ± 16% and over-reported physical activity by 51 ± 75%. Their measured total energy expenditure and resting metabolic rate sat within 5% of the values predicted for their body composition.

That last sentence is the finding. The metabolism was normal. The measurement was the failure. As the authors put it, the failure to lose weight while eating a diet reported as low in calories was due to an energy intake substantially higher than reported and an overestimation of physical activity, not to an abnormality in thermogenesis.

Two constraints on quoting it. The 47% is a case-series extreme in ten self-selected people, and it must never be presented as what obese people, or athletes, or anyone else does on average. And the ±75% standard deviation on the activity figure is larger than the estimate itself, which is what an n of ten buys you.

What transfers is the structure, not the magnitude. A fighter's app has two inputs — food in, training out — and the published error on both runs the same way: intake down, expenditure up. The deficit on the screen is therefore biased optimistic from both ends simultaneously, and the two biases do not cancel. They compound.

The mirror-image failure: manufacturing a problem that is not there

Under-reporting is usually discussed as the reason a cut stalls. It also does the opposite, and the best recent demonstration is in athletes.

McHaffie and colleagues studied 45 adolescent female international soccer players, 16 ± 1 years, during a 9–10 day national-team camp. Self-report used the remote food photography method over three days; doubly labelled water covered seven to eight days, accounting for body-mass change. Self-report gave 2047 ± 383 kcal/day; the isotope-derived figure was 2545 ± 518 kcal/day. That is a mean daily difference of 499 ± 526 kcal and a 22% error. Total daily energy expenditure was 2683 ± 324 kcal/day, or 60 ± 7 kcal per kg of fat-free mass.

Then the consequence. Estimated energy availability by doubly labelled water was 48 ± 14 kcal/kg fat-free mass, range 22–82. By the self-report method it was 37 ± 8, range 22–54. The prevalence of low energy availability below 30 kcal/kg fat-free mass came out at 5% by the isotope method and 15% by self-report — three times as many players flagged as deficient. The authors' own conclusion is that self-report methods can cause a misrepresentation and an over-prevalence of low energy availability, which is the underlying aetiology of relative energy deficiency in sport.

Both failure modes come out of the same bias. A log that misses 500 kcal makes a stalled descent look inexplicable, and it makes an adequately fed athlete look under-fuelled. For an article on the condition itself rather than the measurement of it, see low energy availability in combat sports.

Two limits on this cohort. It is adolescent, female and soccer — the closest athlete data to a fight camp that exists, which says more about the state of the literature than about its applicability. And low energy availability here is a calculated quantity, not a clinical diagnosis.

Where the error actually comes from: portions

The mechanism is not dishonesty. It is that human beings are measurably bad at judging how much food is on a plate, and the badness has a shape.

Nelson, Atkinson and Darbyshire tested this properly in 1994. Fifty-one male and female volunteers aged 18 to 90, from a wide range of social and occupational backgrounds, made 7284 assessments: six portion sizes of each of six foods — two small, two medium, two large — with the actual weighed portion physically on the plate in front of them, estimated against photographs spanning the 5th to 95th centile of the British Adult Dietary Survey.

Using a series of eight graded photographs, mean differences between the portion presented and the estimate ran from −8 to +6 g, or −4% to +5%. Using a single average photograph, they ran from −34 to −1 g, or −23% to +9%. And the direction was not uniform across portion size: large portions tended to be underestimated more than medium or small ones, especially with the average photograph, from −79 to −14 g, or −37% to −13%.

That is regression toward the middle, measured rather than asserted. Small servings drift up, large servings drift down, and the large-portion end is the one that matters to somebody eating in a deficit and eyeballing the rice. The same study found a body mass index of 30 kg/m² or above was associated with an 8% underestimate of portion size, while being female, aged 65 or over, retired, or seeing the photographs in colour were each associated with small but statistically significant overestimations. Note what that sex finding is: a perceptual effect measured against a weighed plate. It is a different construct from the misreporting findings above, points the opposite way, and the two must not be merged.

The estimation aid matters more than the format. Lucassen and colleagues gave 40 Dutch adults (47.5% male, mean age 46.9 ± 19.2 years) one ad libitum lunch with true intake weighed, then collected self-reported portions at two hours and 24 hours in a cross-over design. Reported intakes within 10% of true intake: 31% for text-based estimation using household measures, standard portions or grams, versus 13% for image-based estimation. Median relative error was 0% text-based and 6% image-based. Liquids were the disaster category — 15% median relative error text-based versus 118% image-based. Single-unit foods were the best — 95% of text-based estimates within 10% of true intake versus 29% image-based.

Two other aids have been measured. Gibson and colleagues tested 67 Sydney university staff and students (70% female, mean age 32.7 ± 13.7 years, mean BMI 23.2 ± 3.5, neither overweight nor obese): for geometrically shaped foods, the finger-width method put 80% of estimates within ±25% of true weight and 13% within ±10%, against 29% within ±25% for household measures. Foods that did not conform to a geometric shape — fish fillets, chicken, beef steaks — were overestimated by both methods. And Zoungrana and colleagues, working with 67 mothers and 68 children in Ouagadougou across 1156 paired measurements against weighed records, found a food-photography atlas gave relative errors of 17–56% in women and 25–65% in young children, while salted food replicas gave 21–69%.

Every one of those cohorts is general-population. Nobody has measured how wrong a fighter's portion estimate is, and fight camp — restrained eating, hunger, a scale-anchored goal — is exactly the condition under which perceptual bias would plausibly be worst.

The food's own number is not the food

Even a perfectly honest, perfectly weighed entry inherits whatever error sits in the figure attached to that food.

Urban and colleagues bought 269 food items (242 unique) from 42 restaurants across Massachusetts, Arkansas and Indiana, quick-serve and sit-down, and put them through bomb calorimetry. The headline is reassuring: measured minus stated energy overall was +10 kcal per portion (95% CI −15 to +34; P = .52), no significant average error. The tail is not. Fifty of the 269 items — 19% — measured at least 100 kcal per portion more than stated. And when the 13 worst offenders were re-purchased and re-analysed, the excess reproduced: +289 kcal per portion (95% CI 186–392) first time, +258 (95% CI 154–361) on re-analysis. That was not sampling noise.

A separate study by the same group looked specifically at reduced-energy foods — the ones an athlete in a deficit is most likely to select. Twenty-nine quick-serve and sit-down reduced-energy restaurant foods averaged 18% more energy than stated, and ten supermarket frozen meals averaged 8% more. Some individual restaurant items reached 200% of stated. Free side dishes pushed the energy actually provided to an average of 245% of the stated value for the entrées they accompanied. The authors noted the differences substantially exceeded laboratory measurement error but did not achieve statistical significance, given the variability. Do not turn that into "restaurant calorie counts are wrong by 18%" — it is the reduced-energy subset of 29 items, and the 269-item sample above found no significant mean error at all.

The most recent data is English. Finlay and colleagues calorimetered 295 menu items from 60 outlets in Liverpool and Milton Keynes in 2024, all covered by England's mandatory calorie-labelling policy. Mean reported-minus-measured difference was −16.70 ± 149.19 kcal; mean absolute percentage difference was 21% ± 29%; and 35% of items fell outside the permitted 20% tolerance — 23% overestimated, 11% underestimated by more than 20%. Pubs were worst, at 46% outside tolerance. Matched items across chain locations agreed reasonably well, at 73% within 20% of the paired measurement.

There is one more layer, and it is the database rather than the food. Evenepoel and colleagues compared 100 four-day dietary records from 50 Belgian participants (78% women, mean age 28.2 years) against manually calculated values from the Belgian national food-composition database. For energy the app tracked well: r = 0.96, mean difference +1.3%. Carbohydrate came out −6.4%, fat −1.7%, protein −7.8%, fibre −20% with a fixed bias of about 4 g/day, and cholesterol and sodium were severely underestimated (ρ = 0.51 and 0.53). All correlations at P ≤ .001. The reference standard there is a database, not chemical analysis, so it bounds database error, not food error.

What a label is legally allowed to be

"Nutrition labels can be 20% wrong" is the version of this that circulates, and it is wrong in both major jurisdictions, in different ways.

In the United States, the rule is 21 CFR 101.9(g), and it is asymmetric rather than a symmetric band. Paragraph (g)(3) sets up two classes: Class I is added nutrients in fortified or fabricated foods, Class II is naturally occurring, indigenous nutrients — and if an exogenous source is also added, the total becomes Class I. Under (g)(4)(i), a Class I vitamin, mineral, protein or dietary fibre must be formulated to be at least equal to the declared value. Under (g)(4)(ii), a Class II vitamin, mineral, protein, total carbohydrate, polyunsaturated or monounsaturated fat, or dietary fibre must be at least 80% of the declared value. And under (g)(5), a food declaring calories, total sugars, added sugars, total fat, saturated fat, trans fat, cholesterol or sodium is misbranded if the composite is more than 20% in excess of the declared value.

Paragraph (g)(6) states the asymmetry outright: reasonable excesses of vitamins, minerals, protein, carbohydrate, fibre, sugar alcohols and unsaturated fats over the labelled amount are acceptable under current good manufacturing practice, as are reasonable deficiencies of calories, sugars, fat, saturated fat, trans fat, cholesterol and sodium under it. The regulation polices too little of one group and too much of the other, and tolerates error in the opposite direction for each.

Three qualifications keep this from being usable as a personal error bar. Both (g)(4)(ii) and (g)(5) end with a proviso that no regulatory action follows from a deviation smaller than the variability generally recognised for the analytical method used, so the effective tolerance is wider than the stated percentage. Under (g)(1)–(2), compliance is judged on a lot — a uniform production run, a common container code, or a day's production — sampled as a composite of 12 subsamples taken one each from 12 randomly chosen shipping cases. The single packet a fighter eats is never what was tested. And under (g)(8), compliance may be provided by use of an FDA-approved database computed to FDA guideline procedures, granted where a clear need is presented, such as raw produce and seafood. A declared value can legally be a calculation that was never analysed.

The European tolerances work differently and are the detail most articles get wrong. The European Commission's December 2012 guidance — still listed as current on the Commission's guidance-documents index as of September 2026 — sets tolerances per nutrient and per concentration band, inclusive of measurement uncertainty. Carbohydrate, sugars, protein and fibre: ±2 g per 100 g below 10 g/100 g, ±20% between 10 and 40 g/100 g, ±8 g above 40 g/100 g. Fat: ±1.5 g, ±20%, ±8 g across the same bands. Saturates and mono/polyunsaturated fats: ±0.8 g below 4 g/100 g, ±20% at or above it. Sodium: ±0.15 g below 0.5 g/100 g, ±20% at or above. Vitamins +50%/−35%; minerals +45%/−35%.

There is no energy row in that table at all. Energy is a calculated value derived from the declared macronutrients, so the macro tolerances are what bound it — there is no separate kcal tolerance to quote. The guidance also states that declared values should be average values based on the manufacturer's analysis, a calculation from known or actual average values of the ingredients, or a calculation from generally established and accepted data; and that the measured value should sit within tolerance for the entire shelf life, without being set at either extreme of the range. The guidance has no formal legal status; it is agreed between the Commission and Member States for official controls.

The expenditure side is worse than the intake side

If the food entries are the part of a log people distrust, the expenditure estimate is the part they should.

O'Neill, Corish and Horner pooled 29 studies covering 1430 athletes (822 female, 608 male) and 100 different resting metabolic rate prediction equations, against indirect calorimetry. Of the eleven equations meeting meta-analysis criteria, five did not differ significantly from measured RMR — Cunningham 1980, Harris-Benedict 1918, Cunningham 1991, De Lorenzo and Ten-Haaf — while Mifflin-St Jeor, Owen, FAO/WHO/UNU, Nelson and Koehler significantly under- or overestimated. Effect sizes ran from 0.04 to −1.49.

Agreement on the group mean is not the number that matters to one person, though, and the precision analysis is where this becomes concrete. Of the nine equations meeting the precision criteria, Ten-Haaf was the most precise, predicting 80.2% of participants to be within ±10% of measured values, with all others ranging from 40.7 to 63.7% — De Lorenzo 63.7%, Cunningham 1980 54.1%, Harris-Benedict 53.7%, Owen 40.7%. In plain terms: with most equations, roughly half of athletes have a resting metabolic rate more than 10% away from what the calculator says. The sport breakdown was mixed sports (8 studies), endurance (5), recreational exercisers (5), rugby (3) and other (8). No combat-sport subgroup was named. The authors also caution against interpreting the ratio of measured to predicted RMR as a proxy for energy availability from a single measurement.

Wrist wearables are the other half. Kostrna and colleagues tested four devices against COSMED K5 indirect calorimetry in 58 Hispanic adults (31 female, mean age 23, mean BMI 29.73) during a controlled recumbent cycling protocol. Absolute percentage errors for energy expenditure: Apple Watch Series 8 35.74% (bias +21.60 kcal), Fitbit Sense 2 31.67% (bias +3.14 kcal after device-specific outlier removal, +128.6 kcal with outliers included), Garmin Forerunner 955 77.56% (bias +68.61 kcal), Samsung Galaxy Watch 5 66.49% (bias +56.76 kcal). Higher body fat percentage was associated with greater absolute percentage error across all devices, with a device-by-body-fat interaction at p = .02; Fitzpatrick skin type showed no robust main effect (p = .89), with the authors flagging limited power for type V.

Recumbent cycling is a steady-state, easy-to-model activity. Nothing found validates a wrist device during sparring, bag work, grappling or a wrestling practice — the exact activities a fighter's app is pricing. Manufacturer-published accuracy figures for active-calorie estimation are vendor material and are not evidence; they are not reproduced here.

A worked scenario: Mara Delgado, four weeks out

Mara Delgado is invented. She is a flyweight, four weeks from a bout, and she exists here only to carry the arithmetic — she is not a client, and Fighter Cut has no coached roster.

Her logged day is unremarkable and honest: oats and whey at breakfast, chicken and rice at lunch, salmon and sweet potato in the evening, yoghurt before bed. She weighs some of it and estimates the rest. The app shows a deficit against an estimated total daily energy expenditure, and projects a descent. Four weeks in, the scale has moved less than the projection said it would. She has logged every single day.

Walk the four error terms, each with its published magnitude and its direction.

The expenditure estimate was wrong before she ate anything. The plan's total daily energy expenditure rests on a predicted resting metabolic rate. In 1430 athletes, the most precise equation put 80.2% within ±10% of measured and every other tested equation managed 40.7–63.7%. If her true resting rate sits below the prediction, a meaningful share of the missing deficit is gone before any logging error at all.

The training calories are the least reliable number on the screen. Wrist devices showed absolute percentage errors of 31.7–77.6% against indirect calorimetry during recumbent cycling, in 58 adults. If those minutes are being added back into an allowance, the error compounds straight into the deficit.

The food entries are biased downward, consistently. Athletes under-report by a mean 19% against doubly labelled water, range 0.4–36%. The mechanism is untracked cooking oil, an estimated rather than weighed portion — and large portions are the ones underestimated most, by −37% to −13% against the weighed plate — and a forgotten evening snack.

Even a perfect entry carries the food's own error. Nineteen per cent of 269 restaurant items measured at least 100 kcal per portion above stated, and in the 2024 England calorimetry study 35% of 295 menu items fell outside the 20% tolerance with a mean absolute difference of 21%.

Put plausible values on those four and an intended deficit shrinks toward something much smaller than planned — which fits "less movement than projected" considerably better than either "the plan is wrong" or "she is cheating". Neither of those is the diagnosis. The measurement is.

What the position stand does and does not give you

The 2025 ISSN position stand on nutrition and weight-cut strategies for mixed martial arts and other combat sports is the reference document most fight-camp plans are built on, and it is worth being precise about what it contains.

It gives macronutrient floors for the weight-descent phase: carbohydrate 3.0–4.0 g/kg, protein 1.2–2.0 g/kg, fat 0.5 to 1.0 g/kg/day. Those are floors during longitudinal weight descent, not targets, and they are a different set from the stand's off-camp figures (carbohydrate 4–5 g/kg, protein 1.2–2.4 g/kg with amounts closer to 2 g/kg as a target, fat 20–35% of daily calories or about 1.0 g/kg/day). The position stand gives macronutrient floors in grams per kilogram and a weekly rate. It does not give a calorie prescription, and this article does not convert its floors into one.

On rate, the stand records that combat athletes are commonly recommended to aim for weight loss of 0.5–1 kg of body mass each week, with the resulting energy deficit being based largely on total daily energy needs. Read that against the previous section: the deficit is anchored to an estimate that is outside ±10% for roughly half of athletes with most equations. The stand also reports minimal performance decrement at weight-loss rates around 0.7% body mass per week, and associates descents of 13–15% of body mass over an eight-week camp with large decreases in testosterone and a fall in resting metabolic rate — reported by the stand as a synthesis of other people's measurements, not as a new one, and in a document written around supervised professional practice.

The stand also describes typical acute body-mass losses at fixed distances from a weigh-in: 6.7% at 72 hours, 5.7% at 48 hours and 4.4% at 24 hours, in the context of professional MMA weighing in roughly 24–36 hours before competition, with a multidisciplinary team present. Those three figures are three answers to the same question asked at three moments — they are not stages, they do not stack, and there is no cumulative 17% anywhere in the document. At a same-day weigh-in, the norm in much amateur boxing, BJJ and muay thai, they describe nothing at all. They appear here only because this article's subject is a live descent measured against a division; a food log has no use for them as a schedule.

For context on how those camp numbers are structured week to week, see fight camp nutrition, week by week.

Why the log is still the best instrument available

Everything above is an argument against trusting the absolute number. None of it is an argument against logging, and the distinction is the whole point.

Start with what the error actually is. Between people, under-reporting bias is enormously variable — 0.4% to 36% in the athlete review. Within a person, that bias is a reasonably stable habit: the same oil goes untracked, the same rice is estimated the same way, the same snack is forgotten. A stable bias cancels out of a difference. A change of −300 logged kcal is therefore far more trustworthy than an absolute logged total of 2400, because the first is a comparison of two measurements made with the same broken ruler and the second is a claim about the world.

That is the property to use. Three consequences follow.

Recalibrate against the scale, not the calculator. Body mass over 7–10 days in identical conditions is a measured variable. Total daily energy expenditure is an estimated one. When the two disagree, the estimate is the one that loses — and the published precision figures say exactly why. The mechanics of reading a trend rather than a day are covered in what to track in a fight camp.

Frequency beats precision. Raber and colleagues systematically reviewed dietary self-monitoring in behavioural weight-loss interventions: of 59 included studies, 18 examined adherence against weight loss, and 12 of those found a significant positive association while six did not. The reported pattern was that those completing at least 80% of expected self-monitoring episodes lost significantly more weight, with weight loss associated with the number of diaries submitted. The predictor is the completeness of the log, not the accuracy of the calorie figures inside it. The review's own stated limitation is that variability in adherence measures and limited analysis of weight loss relative to self-monitoring usage limits understanding of how these methods compare. And the cohorts are adults in weight-loss interventions, not athletes.

Reduce the error where it is cheapest to reduce. The portion literature is unusually actionable in one narrow respect: text-based estimation using household measures, standard portions or grams outperformed image-based estimation in Dutch adults (31% versus 13% of intakes within 10% of true), and countable single-unit foods were estimated far better than poured or heaped ones (95% versus 29% within 10%). Liquids were catastrophic by image. A kitchen scale removes the largest measured error term — the −37% to −13% underestimate on large portions — for the foods that carry most of the energy, and leaves the rest estimated. That is a measurement decision, not a diet.

None of this makes the log an audit. It makes it a consistent instrument with a known bias, which is what almost every useful measurement in sport actually is.

What this evidence cannot tell a fighter

The population limits here are severe enough to state on their own, because scaling around them is not possible.

No doubly-labelled-water study of logging accuracy exists in any combat sport. Every magnitude in this article is borrowed from endurance, team, aesthetic or general-adult cohorts, and the closest athlete data is adolescent female soccer players in a training camp — not fighters in a cut.

Fight camp is the worst possible condition for this evidence to transfer, and nobody has measured it. All the doubly-labelled-water under-reporting figures come from athletes in ordinary training, not from athletes in an energy deficit while restricting fluid. Under-reporting is plausibly worse under restraint, but that is an inference, not a finding.

Women are present but thinly, and inconsistently. The soccer cohort is female-only and adolescent; the RMR meta-analysis is 822 female and 608 male; the Chinese combat cohort is female-only but uses a questionnaire screen. The sex differences in misreporting are method-dependent and inconsistent.

Adolescents appear in one athlete cohort — the 16-year-old soccer players — and nowhere else. High-school wrestlers and junior judoka, the population with the most aggressive weight-cutting norms, are absent from every accuracy dataset here.

Amateurs are absent throughout the athlete literature. Every athlete cohort is national-team, collegiate or professional. The reader who logs on a phone between shifts is not represented — though in the portion-estimation studies, where the cohorts are ordinary adults, they arguably are.

One combat-specific finding is worth naming for what it is not. Liang and colleagues assessed 84 female combat-sport athletes in Beijing — judo, freestyle wrestling and sanda, 42 elite and 42 recreational — using food weighing over three consecutive days and heart-rate-derived training expenditure, and found 45.2% (n = 38) at increased risk of low energy availability by LEAF-Q score of 8 or above, with 21.4% (n = 18) at high eating-disorder risk. LEAF-Q is a questionnaire screen. It is not a measured energy availability and it is not a diagnosis, and it does not validate any logging method.

What we could not verify

The research pass behind this article refused twelve claims outright. Naming them is more useful than quietly omitting them, because most are in circulation.

Two studies that appear not to exist. A "2019 study in Nutrition Journal" reporting that crowdsourced nutrition databases contained errors in 27% of entries, and a "2019 study in the Journal of Nutrition" reporting error rates above 10% in about 25% of scanned items with MyFitnessPal worse than Lose It. Both surfaced only inside AI search summaries. No primary paper could be located for either, at either journal. They are not printed here, and they should not be printed anywhere until someone produces the paper.

The circulating portion-estimation percentages. "−43% for condiments to +156% for pasta", "−8.9% to −18.4% under and +9.5% to +90.9% over", "−29.8% for curry sauce to +34.0% for margarine". The first set traces to a 1996 Journal of the American Dietetic Association paper by Blake and colleagues, whose publisher page returns HTTP 403 and for which only title-and-abstract landing pages are reachable. No primary text was obtained. The other two trace to nothing at all. This article uses Nelson 1994 instead, where the figures came with the cohort and the assessment count.

"Nutrition labels can legally be 20% wrong." Refused in both directions. The EU tolerance table has no energy row and works in per-nutrient, per-concentration bands; the US rule is a floor for one group of nutrients and a ceiling for another, applied to a 12-subsample composite of a production lot with an analytical-variability proviso on top.

"Wearable mean absolute percentage error often exceeds 20%" as a general statement. Wearable error is protocol-dependent. The figures here are device-by-device, against COSMED K5, during recumbent cycling, in 58 adults — and they do not transfer to a sparring round. A pooled Apple Watch figure from a living systematic review in npj Digital Medicine was behind an authentication redirect and could not be read; nothing from it appears here.

"Photo logging is more accurate than typing in a portion." One source contradicts this directly. Lucassen and colleagues found the opposite in adults: 31% of text-based estimates within 10% of true intake against 13% image-based, and a 118% median relative error for liquids by image. That sits in tension with Nelson 1994, where a graded series of eight photographs performed well and a single average photograph did not — which suggests the variable is the design of the aid, not photography as a category. Either way, the confident version of the claim is not supported.

Two older portion papers behind publisher walls. Guthrie 1984 and Gersovitz 1978 are cited throughout this literature, including for the "flat-slope" name itself, but neither could be fetched. Where their figures appear elsewhere they are restatements inside an Institute of Medicine review, which adds its own caveat that the flat-slope pattern may reflect random measurement error rather than a true bias. This article uses the measured Nelson result instead.

Three structural gaps. Nobody has compared a logged home-cooked day against bomb calorimetry — the restaurant studies analyse restaurant food, and the app study compares an app to a database. No portion-estimation study exists in athletes, adolescents or amateurs. And the step from "you are 20–40% wrong on a large portion" to "your daily total is X% wrong" is an inference, not a measured finding: every portion study here is a single-meal or single-item laboratory comparison.

Sources

Sourced to

1. Validity of Dietary Assessment in Athletes: A Systematic Review — Capling L, Beck KL, Gifford JA, Slater G, Flood VM, O'Connor H, Nutrients, 3 December 2017;9(12):1313. DOI 10.3390/nu9121313, PMID 29207495
2. Validity of Dietary Assessment Methods When Compared to the Method of Doubly Labeled Water: A Systematic Review in Adults — Burrows TL, Ho YY, Rollo ME, Collins CE, Frontiers in Endocrinology, 11 December 2019;10:850. DOI 10.3389/fendo.2019.00850
3. Discrepancy between self-reported and actual caloric intake and exercise in obese subjects — Lichtman SW, Pisarska K, Berman ER, et al., New England Journal of Medicine, 31 December 1992;327(27):1893–1898. DOI 10.1056/NEJM199212313272701, PMID 1454084
4. Energy expenditure, intake and availability in female soccer players via doubly labelled water: Are we misrepresenting low energy availability? — McHaffie SJ, Langan-Evans C, Strauss JA, et al., Experimental Physiology, November 2025;110(11):1705–1720. DOI 10.1113/EP091589, PMID 39145767
5. Accuracy of Stated Energy Contents of Restaurant Foods — Urban LE, McCrory MA, Dallal GE, et al., JAMA, 20 July 2011;306(3):287–293. DOI 10.1001/jama.2011.993
6. The accuracy of stated energy contents of reduced-energy, commercially prepared foods — Urban LE, Dallal GE, Robinson LM, Ausman LM, Saltzman E, Roberts SB, Journal of the American Dietetic Association, January 2010;110(1):116–123. DOI 10.1016/j.jada.2009.10.003, PMID 20102837
7. Accuracy of menu calorie labelling in the England out-of-home food sector during 2024: assessment of a national food policy — Finlay A, Jones A, Thorp P, et al., British Journal of Nutrition, 2025;134(8):696–704. DOI 10.1017/S0007114525105217
8. Accuracy of Nutrient Calculations Using the Consumer-Focused Online App MyFitnessPal: Validation Study — Evenepoel C, Clevers E, Deroover L, et al., Journal of Medical Internet Research, 19 October 2020;22(10):e18237. DOI 10.2196/18237, PMID 33084583
9. Accuracy of Resting Metabolic Rate Prediction Equations in Athletes: A Systematic Review with Meta-analysis — O'Neill JER, Corish CA, Horner K, Sports Medicine, December 2023;53(12):2373–2398. DOI 10.1007/s40279-023-01896-z, PMID 37632665
10. Body fat, skin tone, and the accuracy of smartwatch caloric expenditure estimates — Kostrna J, Oparina E, Palacios C, et al., PLOS ONE, 2026;21(7):e0353261. DOI 10.1371/journal.pone.0353261
11. Food Photography I: the perception of food portion size from photographs — Nelson M, Atkinson M, Darbyshire S, British Journal of Nutrition, November 1994;72(5):649–663. DOI 10.1079/BJN19940069
12. The accuracy of portion size estimation using food images and textual descriptions of portion sizes: an evaluation study — Lucassen DA, Willemsen RF, Geelen A, Brouwer-Brolsma EM, Feskens EJM, Journal of Human Nutrition and Dietetics, December 2021;34(6):945–952. DOI 10.1111/jhn.12878, PMID 33761165
13. Estimation of food portion sizes in women of childbearing age and young children in Ouagadougou (Burkina Faso) using a food photography atlas and salted replicas: Comparison with weighed records — Zoungrana S, Somé JW, Martin-Prével Y, Lanou HB, Kouanda S, Mouquet-Rivier C, PLOS ONE, 18 September 2023;18(9):e0291375. DOI 10.1371/journal.pone.0291375, PMID 37721927
14. Accuracy of hands v. household measures as portion size estimation aids — Gibson AA, Hsu MSH, Rangan AM, et al., Journal of Nutritional Science, 2016;5:e29. DOI 10.1017/jns.2016.22
15. A systematic review of the use of dietary self-monitoring in behavioural weight loss interventions: delivery, intensity and effectiveness — Raber M, Liao Y, Rara A, et al., Public Health Nutrition, December 2021;24(17):5885–5913. DOI 10.1017/S136898002100358X, PMID 34412727
16. International Society of Sports Nutrition position stand: nutrition and weight cut strategies for mixed martial arts and other combat sports — Ricci AA, Evans C, Stull C, et al., Journal of the International Society of Sports Nutrition, 9 March 2025;22(1):2467909. DOI 10.1080/15502783.2025.2467909, PMID 40059405
17. The impact of low energy availability risk on pre-competition physiological function in Chinese female combat athletes — Liang Y, Li Y, Chen Y, et al., Journal of the International Society of Sports Nutrition, 2025;22(1):2490170. DOI 10.1080/15502783.2025.2490170, PMID 40254934
18. Code of Federal Regulations, Title 21, § 101.9 — Nutrition labeling of food, paragraph (g) — US Food and Drug Administration, Code of Federal Regulations annual edition 2024, published by the US Government Publishing Office, retrieved 7 September 2026
19. Guidance document for competent authorities for the control of compliance with EU legislation on the setting of tolerances for nutrient values declared on a label — European Commission, Health and Consumers Directorate-General, December 2012
20. Critical evaluation of energy intake using the Goldberg cut-off for energy intake:basal metabolic rate. A practical guide to its calculation, use and limitations — Black AE, International Journal of Obesity, September 2000;24(9):1119–1130. PMID 11033980

Read next