Methodology
How we score the evidence.
Verdicts on this site come from clinical trials and meta-analyses, scored with a methodology you can audit. Below: what we use, what we don't, and where the lines are.
418 articles · 9 systematic reviews · 0 affiliate links
Why this exists
Supplement coverage online is dominated by people selling supplements.
The most-Googled "best magnesium for sleep" articles are written by sites that get paid every time you click "buy now." The most-shared "ashwagandha changed my life" threads come from creators who got the product for free. That's the information environment consumers walk into when they want to know whether a supplement actually works.
We made Suplmnt because there was no consumer-facing surface for the evidence the way researchers see it: trial counts, effect sizes, certainty grades, the populations a finding actually applies to. We score everything the same way and report the null findings as prominently as the positive ones — when 30 trials say a thing doesn't help, that's also useful information.
How we score
Two numbers do the work: does it help and by how much.
Step 1 — Pool the trials
Random-effects meta-analysis
For each supplement-outcome pair, we pool the eligible randomized trials using a random-effects model (DerSimonian-Laird with Hartung-Knapp adjustment for small k). Random-effects is the default because real populations vary — assuming a fixed true effect across studies of insomniacs, athletes, and the elderly is wishful.
We report heterogeneity as I² and τ², and we report a 95% prediction interval — what a future trial in a different population is likely to find. When the prediction interval crosses null, that's flagged in the verdict text.
Step 2 — Grade the certainty
GRADE assessment
The pooled effect doesn't tell you how much to trust the estimate. For that we use GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Each domain can downgrade the certainty by one or two levels.
The result is a four-step certainty rating — high, moderate, low, very low — that you'll see on every review page as a colored traffic-light row. Verdict text reflects the certainty: "helps" when the magnitude is meaningful and certainty is moderate-high; "preliminary" when only one or two studies support the signal; "no clear effect" when null findings are confidently null.
Step 3 — Size the impact
MID-units first, Cohen's d as fallback
The biggest mistake in supplement scoring is conflating statistical significance with clinical meaning. A meta-analysis with 8,000 participants will detect a tiny improvement that no one would notice in real life. To avoid that, our primary impact metric is MID-units: |mean difference| / instrument MCID. MCID = the minimum clinically important difference for that instrument. If the effect is ≥1 MCID, it's a clinically meaningful improvement. If it's 0.3 MCID, statistical significance be damned — that's not what someone with that condition would feel.
MID-units only work when there's a native scale with a published MCID — PSQI for sleep quality, PHQ-9 for depression, mmHg for blood pressure, mg/dL for cholesterol. Where they exist, we use them. Where they don't, we fall back to Cohen's d — small (~0.2), medium (~0.5), large (~0.8) — and label the limits of that interpretation.
See: Johnston et al., BMJ 2019; Cochrane Handbook chapters 15 and 18.
What we don't do
The omissions are the position.
- No affiliate links. Every product mention on this site is data — not a paid placement, not a kickback, not a referral. If we recommend a brand, the only thing we earn is your trust.
- No "studies show" without the studies. Every verdict cites the trials it's built from, and you can click through to PubMed for any of them.
- No miracle framing. Effect sizes have ranges. Verdicts have certainty grades. We don't round up.
- No influencer takes as evidence. Anecdotes are interesting. They are not a basis for a verdict on a 30-trial pool.
- No medical advice. This is research translation. Real decisions about your health belong with your physician.
Found something we should improve? We'd rather know.
Methodology lives or dies on critique. If a verdict on this site doesn't match the literature you know, tell us.
Send feedback