Every figure here was read at its primary source. Each carries a confidence interval where the source reports one, a sample size, the study design, and the population it was measured in — because an effect size without those is not information, it is decoration.
11 of these 12 were wrong in our own corpus before an audit on 2026-07-30. Two would have published as roughly 85–88% mortality reductions; one had its direction inverted. The corrections are printed beside each claim rather than quietly applied — see how these numbers are checked.
Last updated 2026-07-30. Claims lacking a confidence interval or a sample size are excluded from this page — two currently are.
We had this wrong. Our corpus stored 0.65. The paper reports 0.49.
Limitations. Observational. Step count is measured, not assigned, so reverse causation (illness reducing steps) cannot be excluded.
We had this wrong. Our corpus stored 0.3 against a "top quartile" contrast. The paper uses percentile performance bands, and reports 0.20.
Limitations. A clinical referral population, not the general public — everyone in it had a reason to be sent for a treadmill test. This limits generalisability and likely inflates the contrast.
We had this wrong. Our corpus stored 0.55 with no interval at all.
Limitations. The cohort is 97% white and healthier than the UK average — UK Biobank has known volunteer bias. Regularity was measured over a single week.
We had this wrong. Our corpus attached this to a node named "top quartile", which inverted the meaning — as stored it read as "strong grip increases mortality". The hazard ratio is per 5 kg *lower* grip.
Limitations. Grip strength is a marker of general robustness, not a lever on its own. Training grip in isolation is not what this measures.
We had this wrong. Our corpus stored the interval as 0.60–1.20, which crosses 1 and would have read as non-significant.
Limitations. Per-increment, not a threshold. Framing it as ">1.2 m/s is safe" misstates the finding.
We had this wrong. Our corpus stored 0.12 in a field rendered as a risk ratio, which published as an 88% mortality reduction. The real figure is a 19% lower risk.
Limitations. Olive oil consumption tracks a whole dietary and socioeconomic pattern. Adjustment cannot fully separate the oil from the life it comes with.
We had this wrong. Our corpus stored 0.15 in a field rendered as a risk ratio, which published as an 85% reduction.
Limitations. The 25–29 g/day "optimal intake" figure often quoted alongside this comes from a separate analysis in the same paper, not from this estimate.
We had this wrong. Our corpus stored 0.65 with a wider interval.
Limitations. Men only, one region of Finland, and only 201 people in the most-exposed group. There is no non-sauna reference group in this cohort — the comparison is frequent vs infrequent, not sauna vs none.
We had this wrong. Our corpus stored 0.75 with a wider interval and did not record the comparator.
Limitations. Women only. The comparator is never-attenders, which matters: the widely quoted "33% lower" figure is for more-than-weekly attendance, a different contrast.
Limitations. Ecological design: the unit of analysis is a county, not a person, so this cannot be read as an individual-level effect. The realised average gain in the studied areas was about 0.4 years. We do not publish a 95% interval here because the paper does not report one.
We had this wrong. Our corpus stored a bare "+10 years" with a fabricated interval of 8–11. The paper reports no intervals for these figures, and +10 applies only to quitting at 30.
Limitations. Men only, born a century ago, in one profession. The size of the effect is not in doubt; the precision is — which is why no interval is shown.
We had this wrong. Our corpus stored a cardiac-events figure against a cardiovascular-mortality outcome, conflating the two, with an interval that implied significance.
Limitations. Two things most summaries get wrong. First, the 2013 paper was RETRACTED in June 2018 for randomisation irregularities — one site randomised whole households, another at clinic level — and republished the same day with per-protocol exclusions and propensity-score adjustment. The effect survived nearly unchanged. Second, this result is for cardiovascular EVENTS. Cardiovascular MORTALITY specifically did not reach significance (HR 0.80, 95% CI 0.51–1.24). "Mediterranean diet reduces cardiovascular death" overstates what this trial showed.
Reuse anything here with attribution. Were the Blue Zones debunked? applies the same standard to the research this field is named after.
If you operate a residential community that already exists and is occupied, this is the evidence base we use to measure it — see whether your community is a match.