How we rate
Every compound is scored on six independent dimensions. They are never averaged, and this page is generated from the same data the compound pages use, so the scale described here is the scale actually being applied.
Why there is no single number
A single score would be more convenient and it would destroy the information you came for. Cardarine scores 1 for acute toxicity and 5 for long-term risk: taking it once is close to harmless, and the concern is a cancer years later. Phenibut is almost the inverse — 5 for dependence, 2 for long-term risk, because the damage it does is to your life rather than your organs. Average either pair and you get a middling number that tells you nothing about which risk you are taking.
So we publish the profile. Higher always means more of that dimension, with one exception, and it is the one people misread most.
The six dimensions
Dependence liability
Does regular use produce tolerance and physical dependence, and is stopping dangerous.
| 1 / 5 | No meaningful tolerance or physical dependence. |
|---|---|
| 2 / 5 | Mild tolerance; stopping is uncomfortable at most. |
| 3 / 5 | Clear tolerance; psychological dependence documented. |
| 4 / 5 | Physical dependence documented; stopping is difficult and needs planning. |
| 5 / 5 | Physical dependence develops fast; abrupt cessation can be medically dangerous. |
Acute toxicity
Overdose potential, interaction danger, how bad a single mistake can be.
| 1 / 5 | A single excessive amount is not expected to cause serious harm. |
|---|---|
| 2 / 5 | Unpleasant but self-limiting in overdose. |
| 3 / 5 | Overdose produces real symptoms; some interactions matter. |
| 4 / 5 | Overdose commonly reaches medical attention; dangerous in combination. |
| 5 / 5 | A single mistake can be life-threatening, especially combined. |
Documented serious harm
Case reports, hospitalisations, and deaths in humans.
| 1 / 5 | No credible reports of serious harm in humans. |
|---|---|
| 2 / 5 | Isolated reports, causation unclear. |
| 3 / 5 | A consistent set of published case reports. |
| 4 / 5 | Many case reports and hospitalisations, or deaths that occur mainly in combination with other drugs. |
| 5 / 5 | Deaths attributable to the compound itself, or harm frequent enough to be routine in clinical settings. |
Long-term / irreversible risk
Carcinogenicity, organ damage, permanent effects.
| 1 / 5 | No signal of lasting harm across substantial human use. |
|---|---|
| 2 / 5 | No known lasting harm, but the follow-up is short. |
| 3 / 5 | Plausible lasting effects; incomplete human data. |
| 4 / 5 | Documented organ-level or endocrine harm that may not fully reverse. |
| 5 / 5 | Demonstrated irreversible harm, including carcinogenicity. |
Evidence quality
How much is actually known. A high score means well-characterised, NOT safe.
This is the dimension that reads backwards. A high score means the compound is well characterised — that we know what it does. It does not mean it is safe. MK-677 scores 4 here because it was studied properly for two years, and what those studies found is part of why it is not an approved medicine.
| 1 / 5 | Essentially nothing in humans. Animal or in-vitro data only. |
|---|---|
| 2 / 5 | Case reports and uncontrolled series only. |
| 3 / 5 | Small or short human trials, or one substantial one. |
| 4 / 5 | Multiple controlled human trials, though gaps remain. |
| 5 / 5 | Large, replicated, long-duration human evidence. |
Product integrity risk
Mislabelling, contamination, and error introduced by the user measuring it.
| 1 / 5 | Regulated supply chain; independent testing is routine. |
|---|---|
| 2 / 5 | Mostly reliable; occasional quality problems. |
| 3 / 5 | Unregulated but usually contains what it claims. |
| 4 / 5 | Analyses regularly find wrong identity, wrong quantity, or contaminants. |
| 5 / 5 | Content is unpredictable, and the user is also measuring or preparing it themselves. |
The five verdict bands
Above the six-dimension profile, each compound carries a plain-language verdict in one of five bands. Bands are distinguished by shape and by name as well as by colour, and never by colour alone.
Well-characterised, low concern (1 of 5)
Studied properly, used widely, and the honest answer is that it is broadly fine. Example: creatine.
Works as advertised, real trade-offs (2 of 5)
It does the thing people take it for, and it costs something specific and known. Example: ostarine.
Largely unknown (3 of 5)
Promising or popular, but the human evidence to judge it does not exist yet. Example: BPC-157.
Serious risk (4 of 5)
Documented harm in humans that is severe, common, or hard to see coming. Example: melanotan II.
The evidence says don’t (5 of 5)
The known downside is severe enough that we will say it plainly rather than hedge. Example: cardarine.
All five bands are in use. That is a deliberate constraint and the build fails if it stops being true. A site where every compound is rated dangerous carries no information, and the first thing a sceptical reader does is look up the one substance they already know about. If we told you creatine was dangerous, you would be right to stop reading.
What would change a rating
- Human trial evidence appearing where there was none. Most of the compounds here are rated partly on absence of data. Data would change that in either direction.
- A documented harm signal becoming a measured rate. Case reports tell you something happens; they cannot tell you how often. A cohort study or a registry analysis would move both the documented-harm and evidence-quality scores.
- A change in the supply chain. Product integrity is about what is actually in the bottle. Regulation, or credible independent testing, would lower those scores.
- Regulatory action — approval, scheduling, withdrawal, or a safety communication from a regulator.
- Us being wrong. See below.
How to challenge a rating
Email [email protected]. What helps most:
- The page and the specific sentence or score you think is wrong.
- What you think it should say instead.
- A citation — a PMID, a DOI, or a link to a regulatory document. We will read it.
You do not need to be a clinician or a researcher. If you took something and our description of what it feels like is wrong, that is worth telling us, and it is the kind of correction that is hardest to get any other way.
Corrections are published in a dated log, not quietly edited into the page. If we got something wrong, the record of having got it wrong stays up.