How to read the research yourself

This site sources every claim so you do not have to take it on trust. That only helps if you can tell a strong source from a weak one. Here is how, in the order that matters.

Not all evidence is the same kind

The single most useful habit is to ask what kind of study a claim rests on, because the kinds are not close to equal. Roughly from strongest to weakest:

Randomised controlled trial strongest: People are randomly assigned to the compound or to a control, ideally without either the participant or the assessor knowing which (double-blind). Randomisation is what lets you say the compound caused the difference rather than merely came before it.
Controlled, not randomised: There is a comparison group, but people were not randomly assigned to it — so the groups may differ in ways that explain the result.
Open-label / uncontrolled: Everyone got the compound and everyone knew it. Improvement here can be the disease running its course, the placebo effect, or wishful measurement. It cannot, on its own, show the compound worked.
Case series / anecdote: A handful of people, or one, reported to have done well. Generates ideas; proves nothing.
Animal or cell-culture weakest for humans: A result in a mouse or a dish. Most compounds that work in a rat do nothing in a person; the graveyard of drug development is full of them. Useful for mechanism, not for effect.

This is exactly the ladder behind the three evidence labels on every entry — preclinical, early-clinical, clinically proven. When an entry describes its evidence, the kind of study is the first thing it names, and it is the first thing you should read.

Five questions that separate strong from weak

1. Was there a control group — and a placebo?

Without a comparison, a result is uninterpretable. People improve on placebo, conditions wax and wane, and someone who paid for and injected a compound is not a neutral judge of whether it worked. A study where everyone got the compound and everyone improved tells you almost nothing. On this site, the epithalamine mortality claim is the textbook case: dramatic numbers, no control group, co-administered with another drug — which is why the entry treats it as unbelievable rather than as evidence.

2. How many people, and for how long?

A trial of nine people can miss a real effect or invent a false one by chance alone. Ask the number (often written n) and the duration. "Clinically studied" attached to a single uncontrolled study in twenty people is doing a lot of work it cannot support. A drug that reduces heart attacks was shown to in seventeen thousand people over years; that is what a strong human result looks like.

3. Did it measure something that matters, or a stand-in?

A surrogate endpoint is a lab number assumed to predict a real outcome — and the assumption often fails. "Raises NAD⁺ in the blood" is not "makes you live longer" or "improves anything you would notice"; the NAD⁺ precursors on this site reliably do the first and, in the trials, not the second. "Builds bone density" is closer to "prevents fractures" than "changes a marker", but they are still different claims. When a study reports a surrogate, ask what it was a surrogate for, and whether that was measured too.

4. Who paid for it, and who ran it?

A manufacturer's own study of its own ingredient is not independent evidence, and much of the cosmetic-peptide "data" is exactly that — a supplier's unpublished panel study on the material it sells. It is not worthless, but it is the interested party marking its own homework, and an honest entry says so. Watch for the conflict-of-interest declaration; its absence is not reassurance, and its presence is not disqualifying, but it changes how much weight a single positive result should carry.

5. Has anyone else reproduced it — or has it been withdrawn?

A finding that no independent group has repeated is provisional, however exciting. And papers are sometimes retracted — withdrawn by the journal for error or misconduct — while continuing to circulate as if they still stood. PDRN's most-cited trial is the example on this site: retracted in 2016 for an undisclosed conflict, yet still quoted everywhere as that compound's strongest evidence. A claim's best study having been retracted is not a detail; it can be the whole story.

Two numbers people misread

Statistical significance is not size. A "significant" result (often p below 0.05) means the effect is probably not pure chance — not that it is large or that it matters. A trial can be significant and clinically trivial: a real effect of half a kilogram, or two-tenths of a percent. Always read the actual effect size next to the significance, and ask whether a person would notice it.

Relative change hides absolute change. "Halves the risk" sounds enormous; if the risk went from two in a thousand to one in a thousand, the halving is real and the benefit to you is one in a thousand. Prefer the absolute numbers when you can find them.

Where to read the source itself

Every superscript number on this site is a link to the study behind the claim — click it and read at least the abstract. Beyond that:

PubMed indexes the biomedical literature; search the compound and read what actually exists.
ClinicalTrials.gov lists registered human trials, their design, size and — crucially — whether they finished or were stopped, and sometimes their results.
A systematic review or meta-analysis, when one exists, has already done the work of gathering and weighing every study — the strongest single thing to read, provided it is recent.
EMA and Drugs@FDA publish the assessment behind an approved medicine — the regulator's own reading of the trials.

The honest summary of most compounds on this site is not that they are proven or disproven, but that the evidence is thin — small, uncontrolled, animal-only, or absent. That is not a gap in the writing; it is the finding. Learning to see it is the point.

The basics · Glossary of terms · How entries are written and sourced