How we decide what counts as evidence

Each article gets one label, shown as a bar at the top and again at the bottom. The label covers the main claim of the article, not every sentence in it.

These are our rules. No other body uses exactly this scale, and a different editor could draw the lines in another place. We keep them fixed so that the same kind of evidence always gets the same label here.

The idea is not new. In medicine, GRADE is a widely used system for rating how much confidence a body of evidence deserves. It rates a whole set of studies together, not a single study.² Randomised trials start high, observational studies start low, and the rating can go down for problems such as imprecise results, results that disagree, or studies that do not match the question.² It was built so that guideline writers would stop rating evidence in different ways.¹

We do not use GRADE as written. It has four levels of its own (high, moderate, low, very low) and a long checklist, made for clinical questions. Our labels borrow the logic: look at the kind of studies, check that they agree, and ask whether they match the question. Then we put the answer in plain words.

Here are four claims, each with the label we would give it and the reason. Two are real examples from our own articles. Two are made up to show the extremes.

Two things stand out. First, a label can be lower than the idea behind it. The principle in the second row is well supported, but the exact method is not, so the label is Moderate. Second, Expert opinion is not an insult. A lot of useful coaching practice has never been tested, and it is more honest to say so than to leave it unlabelled.

Look at the label first, then at the note under it. The note says what kind of studies sit behind the claim, or that it is coaching practice. That is the part that tells you how far to lean on it.

• The sources are linked so you can check us. Each one is opened and checked for authors, year, journal and whether it supports the sentence it is attached to.
• The date in the byline is when we last checked the article. A newer study can change a label.
• A label is not a verdict on you. It describes studies of groups of people. How you respond can differ.

None of this is medical advice. If you have pain, an injury or a health condition, talk to a healthcare professional.

We say so. If there are few studies, or none that match the question, the label is Weak or Expert opinion and the note says why. We do not stretch a single study into a rule, and we do not fill the gap with confident wording.

We still need an answer to most practical questions, such as how long to rest. In that case the article gives a cautious default and says that it is a default. Where a range is all the studies support, we give the range instead of a single number.

We show both sides and keep the label modest. In GRADE, results that differ for no clear reason lower the confidence in a body of evidence, especially when some studies show a benefit and others show nothing.³ We follow the same logic.

A real example is how well people judge their reps in reserve. One study of 141 people found a tendency to be more accurate with more training experience.⁴ A later review of 12 studies and 414 people found no clear effect of training status, and large differences between studies.⁵ So in our article on RIR we do not say that experienced lifters are better at it. We say the studies disagree.