# Enough: evidence and claim audit

Audited October 5, 2026. This report replaces the earlier exact-cutoff presentation. The live page has 12 topics. Each has a declared population, outcome, primary source, numerical extraction and limits in `js/enough-data.js`.

## Search and selection

This was a broad source and claim audit, not a registered systematic review or a claim to have read every study ever published. Searches covered protein intake and lean mass, training volume and hypertrophy, steps and mortality, exercise minutes, sleep duration, fruit and vegetables, fiber, alcohol, low-intensity smoking, income and feelings, working hours and output, and retirement assumptions. Searches combined each topic with systematic review, meta-analysis, dose-response, randomized trial, and recent publication years, followed by reference checks and original tables or author-deposited data.

Priority went to relevant systematic reviews and direct assigned-dose trials. Newness alone did not determine selection. Reviews often reuse the same trials or cohorts, so their participant totals must not be added. Recommendations, randomized treatment effects, observational associations, study-arm regressions and arithmetic scenarios answer different questions. They are kept separate.

Sources were checked through publishers, PubMed/PMC, author repositories and deposited data. Protein and training review full texts were also inspected using NCBI BioC XML; older numerical extracts remain documented with their source locations. Some recent papers were accessible only through abstracts. Those support contextual findings, not newly invented chart coordinates. No independent human reviewer or exhaustive database screening certifies this follow-up.

## Protein: a published study curve, without a forced cutoff

The old chart turned a model across study arms into predicted extra pounds of lean mass at each daily intake. Its sharp bend near 0.73 g/lb/day was not an observed biological switch. Changing the drawing to a smooth curve would have concealed the same problem.

[Morton 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC5867436/) pooled 49 trials with 1,863 participants. Added protein produced an average 0.30 kg of additional fat-free mass gain (95% interval 0.09 to 0.52 kg). Its separate dose analysis compared 42 study arms with 723 participants. A segmented model imposed a flat second segment; the estimated bend was 1.62 g/kg/day, with a broad 1.03 to 2.20 interval. The segmented model did not clearly outperform a straight line (p=0.079). The old 2.78 lb chart span was not the randomized supplementation effect and could not forecast an individual’s gain.

[Nunes 2022](https://pmc.ncbi.nlm.nih.gov/articles/PMC8978023/) included 74 randomized trials overall. For lean mass during resistance exercise, 62 trials and 87 intervention groups gave a small additional effect from protein: standardized mean difference 0.22 (0.14 to 0.30), with moderate certainty. Intake subgroups were below 1.2, 1.2 to 1.59 and at least 1.6 g/kg/day. They compared different studies and populations, rather than randomly assigning everyone to those three intakes. They cannot establish a precise daily ceiling. Lean mass includes water and other non-fat tissue.

[Tagawa 2021](https://pmc.ncbi.nlm.nih.gov/articles/PMC7727026/) analyzed 105 articles with mixed populations and training conditions. Its adjusted spline had a different shape, with slower changes above about 1.3 g/kg/day rather than Morton’s hard stop. Only four articles compared multiple supplementation amounts under identical conditions. The spline’s study-arm associations are not a clean intervention effect from adding a particular daily dose.

[Bagheri 2023](https://pubmed.ncbi.nlm.nih.gov/37516903/) directly assigned 1.6 versus 3.2 g/kg/day during 16 weeks of resistance or concurrent training. There were 48 randomized trained young men and 44 completers. It reported no significant lean-mass advantage from the higher intake. This small trial does not prove equivalence, locate a minimum effective intake or establish a universal ceiling. A 2024 association analysis of these same participants is not a second independent trial.

[Li 2026](https://www.frontiersin.org/journals/nutrition/articles/10.3389/fnut.2026.1860234/full) used 36 trials and 1,552 participants. Its dose axis is supplemental grams/day, not total intake per pound. Exploratory supplement-specific plateaus cannot replace the total-intake curve. [Tagawa 2025](https://pubmed.ncbi.nlm.nih.gov/40914512/) estimates requirements using nitrogen balance and amino-acid oxidation, not measured hypertrophy. It was excluded as a source of growth coordinates. A small 2026 three-dose trial in older women and very small high-protein feasibility studies were screened as population-specific evidence, not substituted for the broader review.

The page now plots **Tagawa Figure 2(h)** in the same graph layout as the other topics. It is the fully adjusted resistance-training panel, controlling for age, sex, intervention duration and weight change. It compares lean-body-mass changes across study arms; it does not isolate the causal gain from adding a particular protein dose. Adjustment for weight change also conditions on a potential mediator. Other panels have different shapes and magnitudes, so this curve is named as one study model, not treated as a consensus forecast.

The source figure was inspected at its native 3990 by 2993 resolution through PMC’s image tiles. The centerline and both dashed 95% confidence outlines were traced, calibrated against axis ticks, and sampled at 111 intakes from 0.30 to 1.40 g/lb/day. These are approximate figure readings, inside the visible source range. Short gaps between confidence-outline dashes were linearly interpolated. No curve was refitted or smoothed into a plateau. Source readings are rounded to 0.01 kg before conversion to pounds, and the readout shows one decimal place. The data file preserves original kg values, pixel coordinates, calibration and a source-image checksum. Its pointwise confidence band does not cover individual variation, all confounding or model-selection uncertainty.

The intake recommendation stays separate from the graph and its uncertainty band. The [ISSN position stand](https://pmc.ncbi.nlm.nih.gov/articles/PMC5477153/) gives 1.4 to 2.0 g/kg/day for most exercising individuals. Dividing by 2.20462262185 gives approximately 0.64 to 0.91 g/lb/day, rounded in the headline to about 0.65 to 0.9. Food and supplements count together. The headline is a recommendation, not a confidence interval, a proven minimum, or an upper limit. Energy restriction and clinical conditions can change needs. Moving the graph marker reads this named study model; it does not predict personal muscle growth.

## Every other live topic

| Topic | Evidence checked | Resulting presentation |
|---|---|---|
| Walking | Paluch 2022 age curves, Ding 2025 review and HHS guidelines | One walking card with an age choice. Display 0 to 16,000 steps, retaining source references and available uncertainty. Mark the uncertain near-zero tail; do not invent a zero-step estimate. Use broad flatter regions, not a required 10,000 or an exact stopping point. |
| Exercise | Garcia 2023, Arem 2015, Ekelund 2019 and HHS guidance | Retain deposited source curve; show its confidence band. Headline 150 to 300 moderate minutes/week comes from guidelines, not the former 343-minute calculation. |
| Training sets | Pelland 2026, ACSM 2026 overview and Steele 2026 trial preprint | Retain the dashed model and credible band. Mark the sparse tail above 25 sets. No exact optimum or claim that 45 sets doubles an individual’s growth. |
| Sleep | Yin 2017, Yuan 2024, Chaput 2026 and AASM/SRS consensus | Keep the older curve as a named illustration with confidence shading. Remove the arbitrary 6 to 7.25-hour highlighted margin. Headline uses adult guidance, usually 7 to 9 hours. |
| Fruit and vegetables | Wang 2021 cohort tables and pooled review, Aune 2017 | Keep published cohort groups as dots with intervals. The broad five-serving headline has pooled support, but higher-intake reviews prevent calling five a universal ceiling. |
| Fiber | Reynolds 2019, Yao 2023, newer review context | Replace the old slope with Yao’s larger, updated linear dose summary. Dashed model and slope limits; no exact 32-gram optimum. Do not invent the separate spline. |
| Income | Killingsworth 2021 data, 2023 joint reanalysis and contrary later models | Show deposited mean feeling scores and their own intervals, rather than invalid intervals for a difference from another group. No universal dollar cutoff. |
| Savings | Exact accumulation arithmetic and Bengen’s retirement context | Dashed scenario, assumptions visible, alternate scenarios labeled as assumptions. Remove the arbitrary 73% “enough” headline. |
| Work | Pencavel author model, Collewet/Sauermann 2017 and four-day/stroke context | Historical model is dashed and named. No modern 54-hour target, no substitution of wellbeing or stroke for output. |
| Alcohol | Zhao 2023 and Stockwell 2024 bias reanalysis | Keep signed group estimates and add interval whiskers. No proven low-volume longevity benefit; do not clamp negative central estimates to zero. |
| Smoking | Hackshaw 2018 and the 2025 Cross-Cohort Collaboration | Keep men’s coronary-risk points with intervals. Newer cohort evidence confirms substantial risk at low amounts, without mixing different outcomes into one curve. |

### Walking and exercise

[Paluch 2022](https://discovery.ucl.ac.uk/id/eprint/10170017/) describes flatter mortality associations around 8,000 to 10,000 steps below age 60 and 6,000 to 8,000 at older ages. The plotted tail is preserved, including the younger group’s bend back and wider limits. These are approximate figure readings of observed habits, not assigned walks. Neither a central dip nor a plateau proves an eventual maximum.

The walking display axis now begins at zero. Figure 3 already includes a modeled low-step tail. Its vector strokes begin at about 9 steps for younger adults and 20 for older adults; the extracted curve is not extended beyond those strokes. At zero the control reports no estimate. Source confidence bands are clipped at low counts, so complete intervals are omitted below 3,000 steps in the younger group and 500 in the older group, rather than treating the clipping boundary as a true upper limit. Those near-zero segments are dashed. Where complete bands are visible, estimates use their geometric midpoint on the source log-risk scale. Where clipped, they use the drawn center stroke normalized at its stated reference. All raw figure readings and their original 5,000- and 3,000-step references remain unchanged. The owner requested a comparison with the U.S. average instead. [Tudor-Locke 2009](https://pubmed.ncbi.nlm.nih.gov/19516163/) reports a nationally representative NHANES 2005 to 2006 mean of 6,540 filtered steps/day (SE 106), in 3,744 adults aged 20 or older with at least one valid ten-hour day. The display rounds this dated average to about 6,500 and names the survey years beside both charts. The raw ActiGraph mean was 9,676; the authors filtered low-intensity counts to improve comparability with pedometers. Device definitions and survey years affect the mean, so this is not represented as a current national average. Both age curves now show 100 times (1 minus HR at the selected count divided by HR at 6,500). This is a comparison at a step count, not an estimate of average American mortality risk. The source figure does not provide covariance needed for intervals against this new reference. The original confidence bands are retained in the data but omitted from the display; dividing their bounds by an estimated denominator would not produce valid 95% contrast limits. The displayed risk axis is linear, with signed percentage labels, including “Same risk,” rather than a zero probability. Equal vertical spacing now means equal percentage-point differences. This changes display geometry only, retaining every source point and displayed risk percentage.

[Ding 2025](https://pubmed.ncbi.nlm.nih.gov/40713949/) includes 57 studies across outcomes; its mortality dose analysis uses 14 studies. Seven thousand versus two thousand steps was associated with 47% lower relative mortality risk, with limits corresponding to 40% to 54% lower risk. This supports large early gains, but has a different reference from the age curves. Its numerical table remains supporting evidence, not a second confusing walking section.

[Garcia 2023](https://pmc.ncbi.nlm.nih.gov/articles/PMC10423495/) includes 196 articles covering 94 cohorts and more than 30 million participants across outcomes. The deposited mortality curve gives about 31% lower relative risk at 150 moderate minutes/week and 38.4% at 600, compared with no non-occupational activity. The chart is a fitted observational association. [Arem 2015](https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2212267) and [Ekelund 2019](https://www.bmj.com/content/366/bmj.l4570) use different activity measures and references, so their curves are not stitched together. [HHS guidance](https://odphp.health.gov/our-work/nutrition-physical-activity/physical-activity-guidelines/current-guidelines/top-10-things-know) recommends 150 to 300 moderate minutes and strength training on at least two days, not a mandatory step count.

### Sets and sleep

[Pelland 2026](https://doi.org/10.1007/s40279-025-02344-w) has 35 hypertrophy studies, 220 effects and 1,032 participants. Its deposited contrasts are model comparisons, not an assigned 15-versus-45-set experiment. The plotted percent is a difference in modeled post/pre muscle-size ratios. At 10 versus zero weekly sets the deposited estimate is 4.18%, with a 95% credible interval of 2.49% to 5.81%. If no-training size remains 100 units, the contrast corresponds to about 104.2 with training, not 4.2% of zero growth. The example is visible beside the graph and names the study follow-up, not a weekly growth rate. No universal duration is implied because the trials differ in length. The assumed increasing function and sparse evidence above approximately 25 sets constrain interpretation. The [ACSM 2026 overview](https://pmc.ncbi.nlm.nih.gov/articles/PMC12965823/) synthesizes 137 reviews, with substantial overlap. It supports higher weekly volume, at least 10 sets, and describes diminishing returns; it cannot identify an exact optimum. Its search ends in October 2024. The [Steele 2026 preprint](https://sportrxiv.org/index.php/server/preprint/view/810) compared 9 and 36 fractional sets and found similar outcomes, but indirect muscle-area measures, attrition and analysis amendments limit it. It is identified as a preprint, not decisive proof of a ceiling.

[Yin 2017](https://pmc.ncbi.nlm.nih.gov/articles/PMC5634263/) uses self-reported sleep. [Yuan 2024](https://www.nature.com/articles/s41746-024-01065-0) and [Chaput 2026](https://pubmed.ncbi.nlm.nih.gov/42454954/) use devices and demonstrate how measurement definitions change the curve. Chaput is a UK Biobank cohort study, not a meta-analysis; its mortality sample reaches 76,811 adults. Its different sleep measures give different minima. [AASM/SRS consensus](https://pmc.ncbi.nlm.nih.gov/articles/PMC4442216/) recommends at least seven hours for adults aged 18 to 60. Its [methodology discussion](https://aasm.org/resources/pdf/adultsleepdurationmethods.pdf) supports seven to nine hours and does not establish a universally harmful upper limit. The older chart’s minimum cannot override those recommendations.

### Food

[Wang 2021](https://pmc.ncbi.nlm.nih.gov/articles/PMC8084888/) combines two US cohorts with a 26-study mortality review. Its pooled five-versus-two-serving estimate is 0.87 (0.85 to 0.88), or 13% lower relative risk. The chart retains the exact Nurses Health Study quintiles from Table 2 and now displays their intervals. [Aune 2017](https://pmc.ncbi.nlm.nih.gov/articles/PMC5837313/) reports further associations at higher fruit and vegetable weights. Servings and grams are not interchangeable; neither source establishes a universal stopping point.

[Yao 2023](https://www.frontiersin.org/journals/nutrition/articles/10.3389/fnut.2023.1153165/full) provides the updated plotted fiber slope: 14 studies, 1,367,285 participants and 97,469 deaths in the mortality dose analysis. Each additional 10 g/day has RR 0.90 (0.86 to 0.93), with high heterogeneity. The reconstructed line is 0.90 to the power of (intake minus 7)/10; shading uses the printed slope limits. Seven grams is a display reference, not a measured untreated control. The source’s separate nonlinear spline has larger changes below roughly 15 grams and slower changes above. We do not redraw that spline without numerical coefficients. [Reynolds 2019](https://discovery.dundee.ac.uk/ws/files/30375889/Final_Lancet_for_John.pdf) remains corroborating context. [Ramezani 2024](https://pubmed.ncbi.nlm.nih.gov/38011755/) and [Veronese 2025](https://pubmed.ncbi.nlm.nih.gov/40651334/) were screened as newer review context; accessible aggregate estimates do not supply a better reconstructable daily-dose curve, and overlapping review populations are not added.

### Money and work

[Killingsworth 2021](https://pmc.ncbi.nlm.nih.gov/articles/PMC7848527/) and its [deposited data](https://osf.io/download/cfnbv/) concern employed US app volunteers, aged 18 to 65, in 2009 to 2015 dollars. The chart now plots actual group means on the 100-point feeling scale. Limits are mean plus or minus 1.96 times the published standard error. Subtracting two groups’ lower and upper limits would not produce a valid confidence interval for their difference, so that presentation was removed. The axis explicitly zooms to 55 to 70. The [2023 joint reanalysis](https://pmc.ncbi.nlm.nih.gov/articles/PMC10013834/), [2024 plateau analysis](https://doi.org/10.1016/j.econlet.2024.111730) and [2026 working paper](https://www.fww.ovgu.de/fww_media/femm/femm_2026/2026_02.pdf) differ in models and findings; the working paper is not peer-reviewed confirmation. There is no universal income cutoff, causal prediction or modern inflation-adjusted dollar target.

Savings is an ordinary annuity calculation. With saved fraction s, real annual return r and target withdrawal fraction w, n = log(1 + r(1-s)/(ws))/log(1+r). Start at zero and contribute at year end. The base scenario uses r=0.05 and w=0.04; alternatives use 0.07/0.05 and 0.03/0.03. The curve is the smooth algebraic approximation; with year-end contributions, the readout rounds up to whole annual contributions. Saving 20%, 40% and 50% reaches the target after 37, 22 and 17 contributions respectively. The shaded range is not a confidence interval. [Bengen 1994](https://finalytiq.co.uk/wp-content/uploads/2017/02/Determining-Withdrawal-Rates-Using-Historical-Data.pdf) concerns historical finite retirement periods; it does not validate steady returns or perpetual withdrawals. The current financial research was screened, but no forecast was substituted into this explicitly hypothetical arithmetic.

[Pencavel’s author paper](https://docs.iza.org/dp8129.pdf), Equation 4 and Table 4(ii), gives additional output above 24 hours as 126.089(h-24) minus 4.533 times max(0,h-49) squared. Twenty-four hours is the shortest week in the study’s Table 1 range. The display axis now begins at zero, with no line or estimate below 24. Output still means additional output relative to 24, not total production. It is a model of historical munitions production, not a recommended workweek. Output differs little between 56 and 70 in this model. [Collewet and Sauermann 2017](https://www.sciencedirect.com/science/article/pii/S0927537116302445) gives a later call-centre example of lower hourly productivity during longer days, but no common weekly cutoff. [Four-day-week wellbeing research](https://doi.org/10.1038/s41562-025-02259-6) and [WHO/ILO stroke research](https://doi.org/10.1016/j.envint.2020.105746) measure other outcomes and cannot supply output coordinates.

### Alcohol and smoking

[Zhao 2023](https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2802963) uses a bias-adjusted model of 107 observational cohorts. Low-volume central estimates are below the nondrinker reference, but intervals include no difference. This is uncertainty about benefit, not an exact zero-effect finding. [Stockwell 2024](https://pubmed.ncbi.nlm.nih.gov/38289182/) reanalyzes the same review family: younger cohorts with better abstainer classification give RR 0.98 (0.87 to 1.11) for low-volume drinkers. Those are not 107 additional independent cohorts. Broad exposure groups remain groups, and one [US standard drink](https://www.niaaa.nih.gov/alcohols-effects-health/what-standard-drink) contains 14 g alcohol.

[Hackshaw 2018](https://pmc.ncbi.nlm.nih.gov/articles/PMC5781309/) gives male coronary heart disease risks of 1.48 at one daily cigarette and 2.04 at twenty. The ratio of extra relative risks is 0.48/1.04, about 46%, not half the absolute chance of illness. The [2025 Cross-Cohort Collaboration](https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1004561) covers 22 cohorts and 323,826 adults. Two to five cigarettes/day was associated with all-cause mortality HR 1.60 (1.52 to 1.69). It confirms harm at low amounts, but is a different outcome and mixed-sex population. Its results are supporting evidence, not extra points in the coronary-risk chart.

## Numerical and visual safeguards

Every displayed estimate remains attached to a source and location. Categorical results are dots with interval whiskers and dashed straight joins that guide the eye; there is no fitted response curve between groups. Continuous bands use the source’s own limits when the displayed reference supports them. Walking omits the source-reference band after changing to a national-average comparison, because its new contrast uncertainty is unavailable. Income limits stay in the published score units. Savings alternatives are labeled assumptions. Protein has a lean-mass outcome axis and a named study-model line with its published limits; its intake-guidance headline is separate. Models are dashed, and the sparse sets tail is named and shaded.

All exact 90% markers are removed from the live presentation. They changed when plotting endpoints changed and were not medical, financial or work recommendations. Earlier models and calculations remain in supporting data for reproducibility, explicitly historical. The change does not invent a smooth protein plateau, flat high-set ceiling, certainty band for work, or extrapolation beyond a study’s observed range.

The data validator checks source values, units, intervals, protein figure units, guidance conversion and historical calculations. The browser harness checks direct chart control, touch scrolling, keyboard access, pointer-exit behavior, axes and unclipped intervals at 375, 768 and 1440 pixels, with brief evidence drawers. Those checks do not constitute scientific peer review or physical-device testing.

The group charts now use dashed straight joins through published estimates to share the same visual format. Their key identifies the joins as guides for the eye. Dots, group selection and original 95% interval whiskers remain; no response model or continuous interval between groups is asserted. Walking states the age-specific comparison count directly in each risk readout, while the curve shows only the movable selection. The fixed diamond and Compare label were removed at the owner’s request.

[Antonio 2014](https://link.springer.com/article/10.1186/1550-2783-11-19) tested achieved intakes of 4.4 versus 1.8 g/kg/day (about 2 versus 0.82 g/lb/day). Thirty people completed eight weeks while maintaining their habitual training. No statistically significant fat-free-mass changes over time or between groups were reported. The small sample, dietary self-report and unchanged training limit what follows; lack of significance is not equivalence. This direct trial remains separate supporting evidence, rather than extending the Tagawa Figure 2(h) line beyond its published endpoint near 1.4 g/lb/day.
