How we rate evidence
Every rating in this book is generated by the rules below. They are published so you can check our reasoning, and disagree with it.
A star rating is only useful when its meaning is written down. Each rating here describes where the evidence stands today. As research progresses and new trials report, a rating can rise.
Two axes, never merged
"Does it work?" and "is it safe?" are different questions, and combining them into one score loses information.
- A compound can have strong efficacy evidence and poorly characterised long-term safety: the GLP-1 drugs, on a 20-year horizon.
- A compound can have almost no efficacy evidence and a clean-looking short-term safety record: most of the experimental tier, where "no side effects reported" usually means not yet studied.
So every entry carries two indicators.
Evidence rating: does it work in humans?
| Rating | Criterion |
|---|---|
| ★★★★★ | Approved for the indication in at least one major jurisdiction, supported by multiple phase-3 randomised controlled trials |
| ★★★★☆ | Multiple good-quality human RCTs with consistent results; may be approved for a narrow indication only |
| ★★★☆☆ | At least one adequately powered human RCT, or several small controlled trials, with mixed or preliminary results |
| ★★☆☆☆ | Human data limited to small, uncontrolled or open-label studies, or robust data exists only for an indication other than the one it is sold for |
| ★☆☆☆☆ | No human efficacy trials. Animal, laboratory or mechanistic data only |
The five rules that generate these
1. We rate the most commonly discussed use, not the best-evidenced use. If a compound has decent evidence for indication A and is widely used by consumers for indication B, it is rated on B, and the entry says so plainly. This rule does more work in this book than any other. BPC-157 has its strongest preclinical data in the gut and is most commonly discussed for tendons. Sermorelin's approval was in children with a diagnosed deficiency, and it is widely used by middle-aged adults for body composition.
2. Animal data never lifts a rating above ★☆☆☆☆. It may be described in detail, and it is often where human research begins. It does not count as evidence of efficacy in humans. Most compounds that fall short in human trials looked excellent in mice first.
3. Mechanistic plausibility is not evidence. "It should work because of pathway X" is ★☆☆☆☆. A coherent mechanism shows a hypothesis is worth testing. Testing is what confirms it.
4. Approval for one indication does not rate the compound for another.
5. Negative evidence is still evidence, and it rates higher than none. A compound tested in humans and found not to work is ★★☆☆☆, not ★☆☆☆☆. AOD-9604 did not meet its goals in a 536-person trial. hCG was shown not to work for weight loss in controlled trials. Those trials answered a real question, and volunteers made that possible.
Safety Confidence: how well is it understood?
This rates how much is known, not how safe something is.
| Level | Criterion |
|---|---|
| Well characterised | Large trial populations plus post-marketing surveillance; known profile, known interactions |
| Moderately characterised | Human safety data exists, but populations were small or exposure short |
| Poorly characterised | Limited human safety data; short exposure only |
| Largely unknown | No meaningful human safety data. Absence of reported harm reflects absence of study |
The rule that matters most
"No reported side effects" is never presented as a safety finding where a compound has not been studied.
The phrase is common, and without context it can be misread as a safety finding. Where it is true that few effects have been reported, our entries say:
Few side effects have been reported, but this compound has not been studied in humans at scale. Absence of reports is not evidence of safety.
What the ratings cost us
Our evidence floor means nothing rated ★☆☆☆☆ carries a commercial link in this book, in any country, regardless of demand.
Retatrutide is the most-searched compound in this entire encyclopedia, roughly 2.8 million searches a month, more than semaglutide and tirzepatide combined. It carries no link and earns us nothing, because it is investigational and there is no approved product to point anyone toward.
That is the rule working as designed. It shows our ratings are not for sale.
Independence
No supplier, clinic or telehealth platform sponsors this book, owns it or reviews it before publication. The ratings come from the published rubric applied to the literature; a company cannot buy a star. We do not name suppliers or tell you where to buy unapproved compounds. Where a lawful route exists, usually a prescriber, we say what it is.