How we count evidence.
Most health content states conclusions. These are the rules that decide which conclusions we are willing to state at all — and, more usefully, which ones we throw away. They are why 28 claims are published out of a much larger corpus. See what cleared them.
Support is counted by person, never by source
One expert repeating a claim across a book, five podcasts and a lecture is one supporting voice, not seven. Counting sources is how a single opinion comes to look like a consensus. Where two people share a byline or routinely appear together, that is weak independence and it is collapsed into one voice before anything is counted.
Spoken-only support caps confidence at moderate
However many experts said it on a podcast, if nobody wrote it down and had it reviewed, it does not reach high confidence. Every claim on these pages is currently backed by peer-reviewed sources; spoken sources exist in the corpus but none of them clears the bar for publication on their own.
One person asserting something caps confidence at low — and low is never published
If the strongest thing behind a claim is that a credible person said so, its confidence is low, and low-confidence claims stay in the corpus rather than appearing here. That rule removes a substantial share of what we have extracted. It is meant to.
A conflict of interest is named, not discounted to zero
Someone who sells the thing they are recommending still counts as support — expertise and interest often travel together — but the conflict is recorded against the claim rather than quietly deleted from it.
Disagreement is published, not resolved away
When sources contradict each other, both positions are kept and shown. A page that presents a contested question as settled is more misleading than one that admits the contest, even though it reads as less authoritative.
Conditions travel with the claim
Almost nothing in nutrition or training is true unconditionally. Every claim carries what has to hold for it to apply and who it explicitly does not apply to. A claim quoted without its conditions is a different claim.
Anything that could delay medical care is flagged and routed to a clinician
Claims touching cardiac symptoms, diabetes diagnosis, sleep apnea screening, medication interaction or disordered eating carry a visible safety flag, and the product behind these pages stops coaching entirely on red-flag symptoms.
A human approves every claim before it is public
Extraction is automated; publication is not. Nothing reaches these pages without a person reviewing the claim against its sources and approving it.
The same discipline runs inside the product: totals are computed in code rather than guessed by a model, a photo estimate is shown as a range unless a printed label makes the exact figure knowable, and every number can be traced back to the entries behind it.
This is a summary of published research, not medical advice. It cannot account for your medications, conditions or history. Talk to a clinician before acting on any of it.