Each effect is listed with a verdict, such as "Probably helps (moderate evidence)" or "Makes little or no difference (strong evidence)", and What was measured. The verdict combines the trial result with the evidence grade, which is given in brackets. "Makes little or no difference" with strong or moderate evidence means trials reliably found no effect.
| What trials found | Strong evidence | Moderate evidence | Limited evidence | Very limited evidence |
|---|---|---|---|---|
| A large benefit | Helps a lot (strong evidence) | Probably helps a lot (moderate evidence) | May help a lot (weak evidence) | Unknown (very weak evidence) |
| A moderate benefit | Helps (strong evidence) | Probably helps (moderate evidence) | May help (weak evidence) | Unknown (very weak evidence) |
| A small benefit | Helps a little (strong evidence) | Probably helps a little (moderate evidence) | May help a little (weak evidence) | Unknown (very weak evidence) |
| No benefit | Makes little or no difference (strong evidence) | Probably makes little or no difference (moderate evidence) | May make little or no difference (weak evidence) | Unknown (very weak evidence) |
| A much worse result | Makes it a lot worse (strong evidence) | Probably makes it a lot worse (moderate evidence) | May make it a lot worse (weak evidence) | Unknown (very weak evidence) |
| A worse result | Makes it worse (strong evidence) | Probably makes it worse (moderate evidence) | May make it worse (weak evidence) | Unknown (very weak evidence) |
| A slightly worse result | Makes it a little worse (strong evidence) | Probably makes it a little worse (moderate evidence) | May make it a little worse (weak evidence) | Unknown (very weak evidence) |
| Trials disagree | Trials disagree (strong evidence) | Trials disagree (moderate evidence) | Unclear: trials disagree (weak evidence) | Unclear: trials disagree (weak evidence) |
| Never tested | Unknown (not tested) | Unknown (not tested) | Unknown (not tested) | Unknown (not tested) |
The size words follow the trials: "a lot" for a large effect, nothing extra for a moderate one, and "a little" for a small one. When the trials do not report a clear size, the verdict uses the plain middle form ("helps", "makes it worse"). In verdicts, Limited is written as "weak evidence" and Very limited as "very weak evidence". Side effects use the same words. For example, "Probably no harm (moderate evidence)" means trials found no harm and the evidence is Moderate, and "May happen (weak evidence)" means the harm was reported but the evidence is Limited.
Interactions known only from case reports or label cautions, with no study that measured them, get "Unknown (very weak evidence)".
| Evidence | What it means |
|---|---|
| Strong | Several good trials, or one very large one, agree. New studies are unlikely to change the answer. |
| Moderate | Good trials exist, but one thing holds us back. The answer could still change. |
| Limited | Two problems, such as small trials that disagree. New trials could easily reverse it. |
| Very limited | Three or more problems, or no controlled human trial at all. |
These follow the GRADE levels used in clinical research (high, moderate, low, very low).
Randomized trials in people start at Strong. Each of these problems drops the grade one level:
Rule 5 and the Strong cap don't apply when the claim itself is about the marker. "Omega-3 lowers triglycerides" is about triglycerides, so measuring triglycerides is the outcome. "Omega-3 prevents heart attacks" would need heart attacks to be counted.
| Label | Meaning | Examples |
|---|---|---|
| Hard outcome | Something that happens to people | Deaths, heart attacks, fractures, a diabetes diagnosis |
| Measured | An objective test or reading | Weight, strength, blood pressure, time to fall asleep, infection counts |
| Self-reported | A questionnaire or rating by the participant | Stress scores, mood, soreness |
| Lab marker | A blood, tissue or hormone level | Cortisol, glucose, triglycerides, testosterone |
The label does not change the grade. It shows that, for example, "Moderate, self-reported" and "Moderate, hard outcome" rest on different kinds of data.
The verdict carries the direction of the result. "Trials disagree" needs adequate trials pointing in opposite directions. A single unreplicated trial counts as Limited or Very limited evidence. It does not count as a disagreement.
An effect is listed under What it helps only if its verdict is "Helps" or "Probably helps" (with or without "a lot" or "a little"). No-effect verdicts ("Makes little or no difference", "Probably makes little or no difference" or "May make little or no difference", each with its evidence grade) go under What it doesn't help. Other verdicts stay under Claimed but unsupported or contested. Counts show table rows; anchor ids stay fixed.
The "How it might work" section is not graded. Each theory explains the proposed steps in plain language, then lists the evidence for and against it, saying whether each finding comes from people, animals or lab cells. A mechanism shown in people does not by itself show a benefit; only trials that measure the effect can do that.
A supplement can have strong evidence for one effect and none for five others. One overall grade would average unrelated effects and read as a verdict on the product. Each article instead opens with two lines: the best-supported effect and the most-cited weak claim, each with its own verdict. A short "What you need to know" box follows. It restates the main findings, harms and study doses from that article's own tables and references, and adds nothing new.
Every rating names the population studied, gives a short reason for the grade, and cites at least one source. Observational associations are never described as causes. Harms are graded and cited the same way as benefits.