Our Evidence Grading System
GRADE-aligned, applied per-claim. The same peptide can carry a high rating for one outcome and a low rating for another, because that's how the science actually works.
Want to see this rubric applied? The peptide evidence matrix lists every graded claim in our catalogue side by side, grouped by outcome and sorted by certainty, so you can see exactly where each rating landed and why.
We grade claims, not molecules
Most peptide sites slap a single “evidence level” on each compound. That's misleading. Semaglutide has very strong evidence for weight loss in adults with obesity, and weaker evidence for other proposed uses. BPC-157 has extensive rodent evidence for tissue repair and almost no human evidence for anything. A single sticker can't carry that nuance.
Instead, we grade each specific claim, a defined outcome, in a defined population, at a defined dose, separately. Each claim gets its own tier, its own rationale, and its own citations.
The framework we use is GRADE (Grading of Recommendations Assessment, Development and Evaluation), the same system used by Cochrane, the World Health Organization, the BMJ, and the CDC's Advisory Committee on Immunization Practices. We add one extra tier (“Preclinical Only”) because GRADE itself doesn't cleanly handle peptides with zero human data, which is common in this space.
The five tiers
High Certainty
We are very confident the true effect lies close to the estimate. Well-designed human RCTs with consistent, direct outcomes.
What it requires: ≥1 large well-designed RCT or systematic review of RCTs, consistent results, direct patient-relevant outcomes
Moderate Certainty
The true effect is probably close to the estimate, but it could be substantially different. RCTs with limitations, or strong observational evidence.
What it requires: RCTs with one GRADE downgrade factor, or large/consistent observational human studies
Low Certainty
Confidence is limited; the true effect may be substantially different. Small or short trials, indirect outcomes, or inconsistent results.
What it requires: Small/short trials, surrogate endpoints, or two GRADE downgrade factors
Very Low Certainty
Very little confidence in the effect estimate. Case series, anecdotal reports, or evidence with multiple serious limitations.
What it requires: Case series, n-of-1, biomarker-only endpoints, or three+ GRADE downgrade factors
Preclinical Only
No human clinical evidence. Effect inferred from animal, in-vitro, or mechanistic data, which often does not translate to humans.
What it requires: Rodent or other animal studies, cell-culture work, or mechanism-only reasoning
Why a tier is sometimes lower than you'd expect
A claim doesn't automatically get the highest tier just because RCTs exist. GRADE specifies five factors that downgrade certainty. We list any that apply on every claim card so you can see exactly why a rating landed where it did.
A worked example: semaglutide
Three different claims, three different tiers, same molecule.
Weight loss in adults with obesity (~15% at 68 weeks, 2.4mg/wk)
HighDirect evidence from STEP 1 (NEJM 2021, n=1,961) and STEP 5 (Nature Medicine 2022). Large effect, replicated, no GRADE downgrades apply.
Reducing major cardiovascular events in adults with prior CVD (~20% relative reduction)
HighSELECT trial (NEJM 2023, n=17,604) is large, pre-registered, with a hard composite endpoint. Pending independent replication.
Effect on muscle preservation when paired with resistance training in GLP-1 users
LowBody-composition substudies confirm meaningful lean-mass loss occurs. But direct RCTs comparing with-vs-without resistance training in semaglutide users are limited; most guidance is extrapolated. Downgraded for indirectness and imprecision.
See the full graded claims on the semaglutide page.
How a peptide gets one overall grade
We grade claims, not compounds. A single peptide can have strong evidence for one thing and none at all for another, which is why every peptide page shows a separate tier for each claim.
Some places need one number anyway, including the structured data search engines read. When that happens, a compound's overall grade is the tier of its best-supported claim. Tesamorelin, for example, has a solid randomised trial for visceral fat and a much thinner one for cognition. Averaging those would describe it as weaker than it is, and taking the worst claim would punish it for being honest about its speculative uses. “How good does the evidence get for this compound” is the question a reader is actually asking, and the per-claim table answers the rest.
Where that grade appears as a 1 to 5 rating, the scale is: preclinical 1, very low 2, low 3, moderate 4, high 5. This measures evidence certainty, not product quality. A 1 means nobody has run a human trial yet. It does not mean the compound is harmful, and a 5 does not mean we recommend taking it.
Editorial process
Define the claim
A claim is an outcome, in a defined population, at a defined dose. We don't grade vague statements like “helps with energy.”
Find the evidence
We start with systematic reviews and RCTs in PubMed, Cochrane, and primary journals. Every supporting study is cited with PubMed ID and DOI when available.
Apply GRADE downgrades
For each claim, we assess the five GRADE factors. Any downgrades are listed visibly on the claim card.
Date and review
Every graded claim has a last reviewed date and the reviewer listed. We aim to revisit each peptide's claims at least quarterly, and immediately when major new evidence is published.
Foundational sources for this framework
See it applied to a real peptide.
View graded claims for semaglutide