How We Grade Peptide Evidence
How Peptide Source Book grades peptide evidence: the six levels, where the data comes from (PubMed, ClinicalTrials.gov), and why a human RCT beats animal data.
Most peptide content online flattens everything into “studies show.” We don’t. A finding from a petri dish and a finding from a randomized human trial are not the same kind of evidence, and that difference is the whole point of this site. Here is exactly how we label it and why the distinction matters.
How we grade peptide evidence: the six levels
We assign every study one of six peptide evidence levels. The label appears on each peptide page next to the study it describes.
| Level | What it covers |
|---|---|
| human-rct | Randomized controlled trial in humans. Participants are randomly assigned to treatment or placebo, blinding is applied, and a control group provides the baseline. The strongest design for establishing causation. |
| human-trial | Non-randomized human trial, or an interventional study registered on ClinicalTrials.gov without full RCT design. Phase I safety studies, open-label trials, and observational studies with human subjects fall here. |
| review | Systematic review or meta-analysis that pools and evaluates a body of existing studies. Strong for synthesis, but inherits the weaknesses of the underlying literature. |
| animal | In vivo study conducted in a living animal, typically rat or mouse. The most common level for research peptides. Animal models can establish mechanism of action and safety margins, but human pharmacokinetics often differ substantially. |
| in-vitro | Cell culture or biochemical assay, no living organism. Useful for early mechanism screening; extrapolation to human physiology is highly speculative. |
| unknown | Publication type could not be determined from the record. |
For most research peptides the honest label is animal or in-vitro. The gap between a rat study and a human RCT is not a technicality, it is often the entire body of knowledge that would let a clinician make a confident recommendation. Saying so is not skepticism. It is reading the literature accurately.
Human RCT vs animal study: why the gap matters
The human RCT vs animal study distinction is the most consequential in peptide research. Animal models fail to predict human outcomes more often than the popular science press acknowledges. Absorption, distribution, metabolism, and elimination parameters differ across species. Rodent healing studies frequently use supraphysiological doses relative to body weight that do not translate directly to human dosing. A peptide evidence level of animal should be read as “promising signal, not confirmed in humans,” not as a surrogate for human efficacy.
This is especially important for peptides like BPC-157, where most published literature is animal-based, see the full evidence-profile comparison in BPC-157 vs TB-500.
Where the data comes from
Every study on this site is sourced from primary databases, not secondary blogs or vendor pages.
PubMed / NCBI is the primary research source for indexed studies. PubMed records each article’s publication type, Clinical Trial, Randomized Controlled Trial, Review, In Vitro, and so on, as a structured field, not freetext. When a publication-type field is present, our classification reads it directly. This removes interpretation from the most common case.
ClinicalTrials.gov is the trial registry for interventional studies. Each record carries a structured study type (Interventional vs Observational), phase (Phase I through IV), and current recruitment status (Recruiting, Completed, Terminated, and so on). A study registered as an interventional trial receives a human-trial grade by default; if the ClinicalTrials.gov record specifies a randomized design with masking and allocation, it qualifies for human-rct.
We also draw from Europe PMC for European literature, and from regulatory sources (the Federal Register, EMA, WADA) for legal and approval status. Regulatory data feeds the legal badges on each peptide page, separate from the evidence level, which refers only to study design.
How the grading works
Classification is rule-based first, model-assisted only when the rules cannot decide, and human-reviewed before publication.
Step 1, Structured field lookup. For a PubMed record, the publication type field is read first. For a ClinicalTrials.gov record, the study type and design fields determine the grade. If the fields unambiguously resolve to one of the six levels, the grade is assigned deterministically, no model involvement.
Step 2, Language model triage. Some records have ambiguous publication types, are preprints, or come from sources without structured metadata. For these, a language model reads the abstract and metadata and assigns a provisional grade using a constrained prompt anchored to our six-level schema.
Step 3, Human review. Before any molecule’s evidence summary is published, a human reviewer checks the studies assigned to it, spot-checks the grade against the source record, and confirms the summary represents the full body of literature, not only the most favorable study.
The goal is simple: the peptide evidence level you see next to a study should be defensible if you open the source link yourself.
What we do not grade as evidence
Forum posts, vendor-sponsored white papers, anecdotal testimonials, and unpublished protocols are not assigned a grade. They may appear in a context note explaining what self-experimenters report, but they are clearly separated from the graded research record. The same applies to citations of citations, if the closest source is a blog that paraphrases a paper, we link the paper, not the blog.
See it in practice
The clearest way to see these grades working is on a head-to-head comparison, where two peptides with different evidence profiles are placed side by side. BPC-157 vs TB-500 shows both evidence levels and legal status for each compound.
Legal status is related to evidence but separate. A peptide can have strong animal evidence and still be an unapproved drug in every major jurisdiction. For the regulatory side of this question, see is BPC-157 legal?.
Research information only, not medical advice.
Frequently asked questions
- What are the evidence levels?
- Six: human-rct, human-trial, review, animal, in-vitro, and unknown. Each study is labelled by its strongest design.
- Where does the data come from?
- Primary databases: PubMed, Europe PMC and ClinicalTrials.gov for research and trials, and the Federal Register, EMA and WADA for regulatory status.
- Do you use AI to grade studies?
- Only for records the structured-field rules cannot resolve, and a human reviews every molecule's summary before publication.