This page is for the skeptical reader. It does not ask you to trust the guide — it tries to make the guide checkable, so you can decide for yourself. If something below is vague, that is a bug; tell us (see How to challenge it).
For what the guide covers and what evidence it leans on, see About this guide. This page is about something narrower and more important: how it is made, where judgment enters, and what it deliberately does not do.
The one thing to know first
This guide grades whether claims faithfully match their sources and hang together. It does not grade whether they are right about the world.
Every claim is traced to the study or guideline it came from, and checked that it says what we say it says. That is a real discipline, and most health writing does not do it. But it is not the same as truth. No claim here has been tested by us against a realized outcome — nobody was followed for ten years to see if the advice helped. A page can be internally sound, faithful to every source it cites, and still be wrong about the world, because the underlying science was wrong.
So: a clean, well-sourced page is a sign of care, not a guarantee of correctness. We say this plainly because a guide that let its own tidiness read as proof would be misleading you.
This is not a systematic review
These are structured, source-grounded appraisals — targeted searches plus explicit grading rules — not exhaustive systematic reviews, unless a specific page says otherwise. A systematic review has a prespecified, reproducible, comprehensive search and dual independent screening; we do not claim that bar. We hold ourselves to the same test a reader can apply to anyone: Is This Actually a Systematic Review. Where a claim rests on a narrative review or a single study rather than a systematic one, we are working to flag it as such rather than let it borrow unearned authority.
Who writes it, and what they gain
- Name / responsible person: Johannes Ahlmann
- Relevant background and its limits: n/a
- Funding: None
- Commercial relationships / advertising / affiliate links / product sales: None
- Conflicts of interest, dated: None, 2026-07-31
A single person currently selects the questions, chooses the evidence, interprets it, and writes the conclusions. That concentration is the honest central weakness of the project, and no amount of internal method removes it — see What it deliberately does not do.
How sources are chosen
We prefer the top of the evidence hierarchy first: systematic reviews, meta-analyses and umbrella reviews, and the major guideline and consensus statements built on them — before individual studies. Pooling many studies is harder to fool than any single striking result.
Every source carries a tier — from gold (a systematic review, umbrella review, or guideline with a documented evidence base) down through high, moderate, weak, to mechanism (animal or test-tube). A weak source is not deleted — it is labelled, because in a thin field a clearly-labelled weak source beats silence — but it cannot quietly prop up a confident claim.
An honest caveat about our own history — and what we have since reconstructed. Source selection up to July 2026 was not especially systematic — it grew breadth-first and opportunistically, without a prespecified search for each topic. That much is unchanged, and we do not claim retroactive rigor.
What has changed is transparency. Much of how the back-catalogue was actually found has since been reconstructed, so the trail is substantially documented rather than opaque:
- most sources trace to a recorded search. The large majority came from literature searches (with a smaller set from a separately-recorded research reading-list), and those searches have been recovered and linked back to the sources they produced wherever the record allowed — partially, and labelled as partial, not pretended complete.
- the roads not taken are recorded too — candidates that were surfaced and not acquired are kept with a brief reason (surrogate-only, superseded, redundant, off-scope), so exclusion is visible rather than silent.
- why each source was ingested is recovered for most held sources, from the running work log.
So the position is narrower than a blanket disclaimer: the process was not a systematic review, but it is now largely documented. Transparency is not the same as rigor, and we still claim only the first. What we are actively doing:
- healing what can be healed — re-checking older weak sources against better ones, flagging sources that entered ad-hoc, and identifying topics that have no solid backbone source yet (so we can go find one, or say honestly that none exists);
- disclosing the residual — where the trail is only partial, or a loose selection cannot be repaired, we mark it rather than paper over it.
Going forward, each literature search that feeds a new source is logged at the time — the question, what was searched, and what was included or excluded — so the how-sources-were-found trail is complete from here on, and partially reconstructed behind us.
How to read a claim
The guide is built to let you descend from a fast read to the full evidence trail:
- Provenance is marked. An extracted claim points to its source; the guide’s own reasoning is labelled as such. You can tell what came from a study from what we inferred.
- Four states, kept apart: benefit, harm, no meaningful effect, and not enough evidence to say. The last two are not the same, and we never blur them.
- Words are calibrated (see the table below) — established / probably / may / insufficient evidence map to strength of evidence, not to how confident the writing sounds.
- Certainty is not the same as a recommendation. How sure we are about an effect is a different question from how strongly anyone should act on it — see the machinery in the Evidence Appraisal Hub.
- Every topic page carries an evidence box at the end — the question, what evidence went in (by tier), the overall certainty, a note flagging any weaker sources, and a link to the page’s full public edit history — the “why should I believe this?” layer, with links down into the method behind each field.
The words we use, and what each means
To keep confidence out of the writing style and in an explicit scale, these phrases carry fixed meanings across the guide. They track the four evidence states — benefit, harm, no meaningful effect, insufficient — and the certainty machinery behind them (Rating Certainty of Evidence, The Insufficient-Evidence Statement).
| Phrase | What it means about the evidence |
|---|---|
| Established | Consistent evidence at moderate-to-high certainty |
| Probably | Moderate certainty |
| May / might | Low certainty |
| Insufficient evidence | Very low certainty, or serious gaps — we cannot say yet |
| No meaningful effect | Measured in humans, and a worthwhile effect is adequately excluded |
| Not adequately studied | Absence of evidence, not evidence of absence |
The last two are kept firmly apart: no meaningful effect means we looked and found nothing worth acting on; not adequately studied means the question has not yet been asked well enough to answer.
How to challenge it
Trust is not inferred from the absence of corrections. It is inferred from whether criticism can be submitted, evaluated, and traced to a visible change.
- Report an error or a missing study — open an issue on the public repository: github.com/codinguncut/evidenced.health/issues.
- The full edit history is public. Every change to every page lives in that repository; each topic page also links its own commit history, so any revision can be inspected line by line.
- Substantive challenges are adjudicated against the sources, not by who makes them, and the record of that adjudication is kept.
- As new evidence arrives, older conclusions are re-checked; the public history tracks conclusion-changing revisions, not just typos.
What it deliberately does not do
- It is not independently reviewed. One person makes the calls. Where that matters most, a page says “Not independently reviewed.” We would rather state this than imply a peer-review that did not happen.
- It is not exhaustive. Targeted searches miss things. Absence of a study here is not evidence a study does not exist.
- It holds one axis: health. It does not price environmental, ethical, animal-welfare or cost considerations — it has no data on them and will not pretend to. Where such a trade-off exists, it names that it exists and stops.
- It does not prescribe. It appraises what an exposure does; selecting, dosing and screening for interactions are a clinician’s job, with your labs and history in front of them.
The aim is not to convince you the guide is trustworthy. It is to make how it is produced transparent enough that you can judge, for yourself, whether trust is warranted — and to keep the layers open so a doubter can always dig one level deeper.