How we vet the evidence

Before an article is written, the specific numbers it will use are pulled from primary sources and pinned: the effect size, the confidence interval, the study design, the population. The prose is then written against those pinned figures rather than the other way around.

A figure that cannot be traced to a source in our citation registry does not go in. That is enforced by a build test, not by good intentions, so an article physically cannot ship citing a source that does not exist.

Each drafted claim is checked against the primary publication by independent reviewers whose job is to refute it, not to wave it through. A claim survives only if it holds up against that scrutiny. This is fidelity checking: does the source actually say this number, is the design labeled correctly, is the endpoint right.

That process has caught real errors before publication, including a mislabeled study endpoint and figures that turned out to be model projections rather than measured outcomes. When a widely-repeated number cannot be traced to a primary source, we leave it out and say so.

Every article carries an evidence grade, and the grade is a defined claim rather than a label we like the sound of. It tracks how much pooling and checking already happened upstream, which is why a single simulator study and an umbrella review of dozens of systematic reviews do not carry the same badge.

Most of what is known about shift work comes from observational studies, which can show that two things occur together but cannot prove one caused the other. Those findings are written with associational language. Causal wording is reserved for randomized trials and quasi-experiments, and where a study is correlational we say so plainly, usually in the authors' own words.

Two studies agreeing is not always two independent confirmations. Where every source behind an article comes from a single research program, the article says so, because a shared method or measurement choice would show up in all of them. Agreement within one program is weaker evidence than agreement between separate teams, and we do not let the reader infer the stronger thing.

Some scheduling questions people expect us to answer are not answered here. Where a dedicated search did not surface effect sizes we could verify against primary sources, the topic stays empty until it can be filled properly, and the relevant pages say which topics those are. An empty section is more honest than a confident paragraph with no citation behind it.

None of the studies in this library evaluated Soon, and none of the research was funded by us. These are independent studies of scheduling practices, shift patterns, and working hours. That is a stronger position than a vendor case study, but only if it is stated rather than assumed.