How we check these numbers

Every figure in this library is tied to the sentence it came from. Not paraphrased, not summarised — the actual sentence in the paper’s abstract. That sentence is printed next to the number so you can check it yourself instead of trusting us.

The check that runs

A script re-fetches each source abstract from Europe PMC and asserts two things for every one of the 127 figures across 14 claims: that the value appears insideits own quoted sentence, and that the sentence appears verbatim in that paper’s abstract. It runs on every change and again weekly. If a single number fails, nothing ships.

The weekly run is not decoration. A claim that was true when it was written can be retracted or superseded later, and only a scheduled re-check notices.

What it guarantees

That the number is real, that it is attached to the finding it actually describes, and that nobody transplanted a correct figure onto the wrong outcome. That last one is the failure this was built to stop: a genuine sentence paired with a genuine number from somewhere else reads perfectly and is entirely wrong.

What it does not guarantee

It does not tell you the study was well conducted, that its conclusion is correct, or that it will still look right in five years. A machine can confirm a number was faithfully copied. It cannot referee a field.

So we do two other things. We only publish from meta-analyses and systematic reviews, never single studies. And every page states what the paper does not support, above the numbers rather than below them — because a finding without its limits is how good research turns into bad advice.

On study design

Each page names the design as the abstract establishes it, and no higher. Where a paper does not state that the studies it pooled were randomised, we do not call it a meta-analysis of randomised trials — even where that is the likely case. Not one claim currently in this library survived that check as a meta-analysis of RCTs.

Quoting

Findings, effect sizes and numbers are facts, and we restate and cite them from any paper. Abstract textis usually the publisher’s, so where a paper’s licence does not permit reproducing it, the page shows the numbers and the citation without the quoted sentence, and says so on the page.

Corrections

If you find something wrong here, tell us and we will fix it and say what changed. Getting one of these wrong costs more than never having published it.