Editorial policy

We do not cite papers from memory. Only what we opened and checked on PubMed is written down, and every page lists its sources.

Secondary material does not count as evidence — blogs, press releases, manufacturer documents, other summary sites. Reading those goes only as far as deciding what to look up next.

Where it matters whether we read the abstract or the full text, we say which.

How we look is written down too. We search with PubMed’s API, fetch and read the abstract, and register the paper by its PMID. Retracted papers are filtered out by the registration script.

And every paper carries how far we read it. The default is the abstract; only papers whose full text we opened are marked otherwise. Setting the default the other way would let an untouched entry quietly claim a full-text reading, so where it errs, it errs on the weaker side. An abstract holds only the numbers the authors chose; how many people dropped out, and what was done with the secondary endpoints, are in the full text.

There are four badges — strong, moderate, early, not established.

'Strong evidence' has to satisfy all of this.

• two or more separate randomised controlled trials in humans (or one trial of 1,000+ people over 4+ years with a clinical event as its endpoint)
• a trial that only looked at self-rated improvement cannot be raised to it
• unless the endpoint is clinical, no no-treatment control in the mix — plenty of measures rise from moisturiser alone
• not all of the evidence may be industry-funded

Several papers from one trial are counted as one. A meta-analysis is counted as the number of trials inside it.

A 'strong' badge that breaks these conditions fails the build. It is a check rather than a habit, because people forget.

The conditions count designs. They cannot count what was inside the capsule.

Lactoferrin is the example. Two placebo-controlled randomised trials, a lesion-count endpoint, not all industry money — it passes. But the large trial put three things in one capsule, and the trial that isolated lactoferrin had 36 people. So we stopped at moderate.

It does not work the other way. We never raise a badge by hand past conditions it does not meet.

Results that found no effect are not pushed to the bottom. If one of three trials is negative, that one stays.

Nor do we delete any of this.

• a primary endpoint that only just cleared — we write the lower bound of the interval
• the placebo arm improving too
• sentences where the authors wrote their own limitation — 'blinding was not feasible', 'recurrence could not be assessed', 'better trials are needed'
• the adverse effects of a drug with strong efficacy evidence — written at the same size

We do not estimate concentrations or doses. If the paper does not give one, it stays 'unreported'.

p values and confidence intervals are carried over as published. We do not turn 0.054 into 'almost significant'. It was not significant.

We do not convert a relative risk into absolute numbers. If the abstract gives no incidence, we cannot write 'so many in so many'.

Every study is marked industry, public or unclear.

If an author's affiliation is a manufacturer or its research and development arm, it is industry. A declared royalty or piece of intellectual property counts the same way.

If no funder is stated, it is unclear. A university affiliation alone does not make it public.

The procedure and surgery pages do not deal in amounts of money, named clinics, or treatment testimonials.

Instead there is a relative cost band (1–5) among the procedures on this site. And the field's type is narrowed to 1 through 5 so that a currency amount cannot get in even by accident.

Forbidden phrasing is caught by a check — amounts, clinic recommendations and comparisons, treatment reviews, 'guaranteed results', 'no side effects'.

No stars, no ranking, no 'recommended'. Products are examples; the evidence sits with the ingredient.

And only products verifiable in the Korean functional-cosmetic filing register are listed. Where a content figure is not published, the field stays empty.

We do not fix mistakes quietly. Three times a concern was filed in the wrong slot and corrected later — bruising, leg veins and cellulite. Each page records how it happened.

If you find something wrong, tell us. Send the paper with it and we will fix it and keep the record.

And there is no byline. With no medical qualification or clinical experience to put forward, a name on these pages would carry weight it does not have. In its place stand four things: the source list with PMIDs, how far each paper was read, the date it was last checked, and this record of corrections. For anyone wanting to check, those four are worth more than a name.

This site does not diagnose and does not prescribe. Whether to take a medicine or have a procedure or an operation is settled in a clinic; what this site records is what has been established, and how far.

And it is better not to decide from one clinic's account.