Editorial policy

1. Step 1

   Draft

   DRAFT

   Writer or editor

   The prose is written against a study list that already exists. We do not write a conclusion and then go looking for citations for it.

2. Step 2

   Medical review

   IN_MEDICAL_REVIEW

   Named medical reviewer

   A credentialled clinician checks the claims against the cited studies, sets or confirms the grade for every intervention–outcome pair on the page, and signs it with their name and the date. A page cannot leave this state without a reviewer attached.

3. Step 3

   SEO review

   IN_SEO_REVIEW

   Editor

   Titles, descriptions, internal links, structured data and the verdict block. This step can change how a page is found. It cannot change what the page says about the evidence, and it cannot touch a grade.

4. Step 4

   Scheduled

   SCHEDULED

   Editor

   Approved and queued for a publication time. Nothing in this state is publicly reachable.

5. Step 5

   Published

   PUBLISHED

   System

   Live, with the reviewer byline, the review date and the next review date visible on the page itself.

6. Step 6

   Re-review due

   RE_REVIEW_DUE

   Named medical reviewer

   The page stays live and is flagged for re-review. Standard interval 12 months; 6 months once a page carries real traffic. New evidence can pull a page into this state early.

A published health page that has lost its reviewer attribution or its review date does not degrade gracefully — it 404s. That is deliberate: an unsigned health page that still looks authoritative is worse than a missing one.

What a grade change has to carry

A grade change is refused unless all of the following hold. These are validator rules, not guidance:

• The actor is a medical reviewer with a linked, named reviewer profile.
• A written justification of at least 80 characters and 12 words, of which at least 8 survive after boilerplate phrases are struck out. “Updated per review” is rejected by the validator, not by a human noticing.
• At least one triggering study, unless the move is to INSUFFICIENT — that move often happens precisely because the citable evidence was withdrawn.
• The justification is published verbatim in the public change log. Reviewers write it knowing it will be read by strangers.

Primary literature, appraised individually

Every study in a grade carries its design, species, participant count, duration, a 0–100 risk-of-bias appraisal, and a reviewer weight. None of those is optional and none of them defaults to the flattering value: an unappraised study is scored as average, not as good, and a study with no reported sample size is assumed small.

Syntheses are preferred, and de-duplicated

A meta-analysis or systematic review counts for more than a single trial because it subsumes a body of work. When a synthesis and its constituent trials are both in a base, reviewers are required to down-weight the primaries. Otherwise the same trials are counted twice and the literature looks larger than it is.

We cite what we read

Every claim on a graded page is attached to a citation with a DOI or PMID where one exists. If we cannot link the claim to a source, the claim comes out of the page.

Retractions are tracked, not quietly dropped

A retracted study contributes zero to the score and stays visible on the page, marked as retracted, with a penalty recorded against the grade. Deleting it would hide the fact that the field produced it.

Press releases, podcasts and manufacturer literature are not sources

They can be a reason to go and look for evidence. They are never evidence. A supplement company’s own white paper does not enter an evidence base.

Mechanism is not evidence

A plausible mechanism never raises a grade on its own. Mechanistic reasoning carries the lowest design weight in the engine, and a base with no human data cannot be banded at all.

A preprint is multiplied by 0.6 — worth roughly three fifths of the same paper after peer review. Not zero, because preprints of large registered trials genuinely move what we know, and pretending otherwise would have us ignoring the best available evidence for months at a time. Not close to 1.0, because the specific failure we are guarding against is a hyped preprint driving an A grade before review catches an error in it.

• Preprints are labelled as preprints everywhere they appear, including in study tables.
• A grade may not rest on preprints alone where the alternative is INSUFFICIENT: the banding floors apply after the discount, so a lone preprint cannot band a claim.
• When a preprint is published, the record is updated and the discount is removed, which can move a grade. That move appears in the change log with the peer-reviewed version named as the trigger.

In the evidence

Conflict of interest is recorded per study and penalised once, at the level of the whole evidence base, starting when conflicted studies pass 50% of the base’s evidence mass. Industry funding does not reliably invalidate a single trial, so penalising each one would be wrong; a literature where most of the weight is conflicted has a known publication-bias problem, so ignoring it would also be wrong.

In our reviewers

• Reviewers declare financial interests — equity, advisory roles, speaker fees, consulting — on joining and annually after that.
• A reviewer does not review an intervention they have a financial interest in. They recuse and another reviewer takes the page.
• Reviewers are paid a fixed fee for review work. No part of any reviewer’s compensation depends on the grade they assign, on traffic, or on whether a clinic in our directory offers the thing they graded.
• Reviewer profiles list credentials, specialties and external identifiers so you can check who they are rather than taking our word for it.

In the business

We sell subscriptions, clinic listings, labelled advertising and data access. None of those can buy a grade, a ranking position or the removal of a review. That is a claim, so we explain the mechanism behind it — including where in the code the separation lives — on how we make money.

• AI is used to draft summaries, extract study metadata and surface new papers. Every AI-drafted item starts flagged as AI-drafted and cannot become publicly visible until a named medical reviewer signs it off.
• AI never sets a grade. Grade writes require a medical reviewer identity; there is no automated path to one.
• Anything generated by a model and shown to you is labelled as such, on the page, next to the output rather than in a footer.

If something on this site is wrong, email editorial@longlifelist.com with the URL and what is wrong. A citation makes it much faster. We aim to respond within five working days, and to correct a factual error within ten of confirming it.

• A typo or a broken link is fixed silently. Anything that changes the meaning of a claim is logged as a correction, with what changed and why, on the page itself and in the public corrections log.
• A change to a grade is logged separately, in the grade change log, with the reviewer’s justification unedited.
• We do not remove a correction notice after the fact. The notice is the record that we got it wrong.

Complaints about a clinic listing, a review, or a safety concern have their own routes on the contact page, because they go to different teams and some of them have legal deadlines.