The Weight of Evidence

Method

Where the claims come from

We scan the comment sections of health podcast channels on YouTube. Not the episodes — the comments. What we look for is a health claim stated specifically enough to be wrong: a mechanism, a number, a threshold, an “X causes / prevents / cures Y”.

A claim that cannot fail any test is not a claim we can adjudicate, and we discard it.

How a claim is chosen

Not by how interesting it is. Two things decide: whether it is falsifiable, and how much it circulates.

Circulation is measured — how many times the claim appears, across how many distinct channels, with what engagement, over what window. Every adjudication on this site publishes those numbers for the claim it examines.

This fixes what we are answering. We are not asking “is walking good for you”. We are asking whether a specific sentence, which a specific number of people are repeating to each other right now, survives contact with the literature.

Whose claim it is

The claims come from comment sections. They are not the position of the podcast, its host, or its guests, and we never present them as such. When we name a channel, we are naming where a claim was circulating — not who made it.

Commenters are never identified. No usernames, no screenshots, no handles. We quote briefly or paraphrase, and we attack the claim, never the person who wrote it.

How the evidence is gathered

We search the primary literature directly, not press releases and not secondary coverage.

One detail worth stating, because it changes results: literature databases return newest-first by default. Searching that way systematically buries the paper a field actually relies on. So every search runs at least twice — once by date, once by citation weight — and then follows the references of what it finds. The canonical paper is usually not the recent one.

How evidence is weighed

Not all evidence counts the same, and we say which is which on every page:

A claim supported only by the bottom of that list is reported as exactly that.

Every claim in this field gets asked two questions, because they are where most of them fail.

Which way does the arrow point? Does the exposure change the outcome, or does the outcome change the exposure?

Does the threshold survive replication? A number that moves between cohorts is a property of the cohort, not of the human body.

What a verdict means

The verdict applies to the claim as stated, not to the topic. A claim can be directionally right and quantitatively wrong, and that is a different verdict from either “true” or “false”. We say which.

Every adjudication states what remains uncertain. If we cannot say, we publish that we cannot say.

We do not issue a verdict on the strength of a single study. A new paper is reported as a new paper: here is what it claims, here is what the prior literature said, here is what would have to be true for it to hold. Speed is for tracking. Verdicts wait.

What is automated and what is not

The scanning, the deduplication, the circulation scoring and the first literature sweep are automated. So is part of the drafting, and the narration in the videos is synthetic.

The selection, the reading, the weighing and the verdict are not. Every page is reviewed by a named human before it publishes, and nothing publishes without that review.

What we do not do

We do not republish podcast audio or transcripts. We do not reproduce figures or tables from papers — where a chart appears here, we rebuilt it from the reported numbers. We take no money from anyone whose product could appear in an adjudication, and there are no affiliate links on this site.