The evidence

Evidence first, argument second.

We are building a structured, public record of what research actually shows — including where it is uncertain, contested, or simply not yet available.

Published research

No research records have been published yet. Every record goes through the review described below before it appears, and we would rather publish nothing than publish something unchecked.

Browse the research library →

What a study says, and what we say

Every record keeps three things apart: what the research found, what limits that finding, and why we think it matters.

The first two describe the source. The third is ours, and it is always labelled as ours, so our reading of a study can never be mistaken for the study itself.

A finding is described as an association unless the study design supports a cause.
Self-reported answers are described as self-reported.
A finding about one group is not presented as true of everyone.
Research that complicates or contradicts our position is eligible for the library on the same terms as research that supports it.

Before anything is published

A record becomes public evidence only when all of these are true:

The original source is identified — not a news story about it.
The source can be accessed, or its identity independently checked.
Its details — title, authors, publication, date — are checked.
The study design is classified by what the study actually did.
The key finding is checked against the original source.
The important limitations are recorded.
Our interpretation is kept separate from the source's finding.
A person who did not write the record reviews it before it is published.

Kinds of research

Each record is labelled with the design the study actually used. We never promote a study to a stronger kind of evidence than it is.

Systematic review or meta-analysis: Combines many studies. We look at the quality of the review and how much the studies disagree.
Randomised or experimental study: Can test cause and effect, when it is well designed.
Longitudinal or cohort study: Follows people over time. Shows what comes before what, but other factors can still explain a link.
Cross-sectional study: A snapshot. Good for how common something is and what goes together; usually cannot say which came first.
Qualitative research: How people experience something and why. Rich in meaning, but not a measure of how common it is.
Laboratory or behavioural experiment: Controlled conditions. We consider how far the results hold outside the lab.
Policy or regulatory report: Context on policy and how it works in practice. An institution's position is not independent research.
Expert commentary: Useful for interpretation, but not the same as evidence from a study.

Sources, AI and corrections

We prefer peer-reviewed research, high-quality reviews, public-health publications, academic institutions and transparent datasets. News articles help us find research; we cite the research itself. We do not use unsourced social posts, press releases, screenshots or AI-generated citations as evidence.

AI tools may help us find and summarise research. They are never the judge of whether a claim is true: every published citation and finding is checked against the original by a person.

When a source is corrected, retracted or withdrawn, we mark it clearly, record when and why, and update what depends on it. Mistakes are corrected openly.

We do not yet label evidence as “strong” or “limited”. We will only do that once the rules for assigning those labels have been written down and reviewed.