How we evaluate evidence.

Every research entry on this site is assigned a single tier from 1 to 4. The tier reflects how directly the source observes the claim it's used to support.

Tier 1 — Peer-reviewed primary research

Randomized controlled trials, mechanistic studies, well-characterized cohort studies. The source is reporting first-hand observation of the phenomenon under controlled conditions, peer-reviewed before publication. Most of the corpus's load-bearing claims for FDA-approved peptides (semaglutide, tirzepatide, tesamorelin, liraglutide, bremelanotide) rest on Tier 1 evidence.

Tier 2 — Peer-reviewed secondary research

Systematic reviews, meta-analyses, narrative reviews from credentialed authors in peer-reviewed journals. Tier 2 entries are useful for synthesis but are downstream of the primary literature they cite — we treat them as orientation, not as terminal evidence.

Tier 3 — Expert primary observation

Named credentialed clinicians and researchers reporting in non-peer-reviewed venues: preprints, institutional reports, conference proceedings, named blog posts from PhDs and MDs. The author identity is verifiable; the methodology may be less rigorous than peer-reviewed primary work.

Tier 4 — Practitioner experience and case reports

Anonymized practitioner observation, individual case reports, aggregated community self-reports. Useful for hypothesis generation and for understanding what the real-world adoption looks like; not sufficient on its own to support a claim.

What we don't do: aggregate multiple Tier 4 sources into a Tier 1 conclusion. Five forum threads do not become a randomized trial. The tier of a claim is constrained by the tier of its strongest supporting source.

Independent of tier, every cited source is labelled with how robustly it supports the specific claim it backs. A Tier 1 RCT can still produce suggestive evidence if the trial is small, has methodological issues, or is the sole data point for a claim. Conversely, a Tier 4 community-report aggregate can produce moderate evidence if the replication is consistent across many independent observations.

• Strong — large, well-designed, well-replicated. Effect direction and magnitude are reliable.
• Moderate — well-designed but small or unreplicated; consistent direction across multiple sources.
• Suggestive — single study, smaller effect, or methodological limitations. Direction plausible; magnitude uncertain.
• Anecdotal — case reports, individual practitioner observation, community self-report. Hypothesis-generating, not evidentially conclusive.

Why two dimensions instead of one? A meta-analysis of methodologically weak studies is still weak; a single well-designed mechanistic study can be strong even though it's Tier 1. Collapsing both into a single "evidence grade" hides the difference.

When the evidence is thin, we say it's thin. When a peptide is mechanistically plausible but lacks human RCTs, we say so. When community use diverges from what the trial data supports, we describe both the trial data and the community pattern, separately, without conflating them.

Specific patterns we won't accept:

• "More research is needed." Vague. We instead state which kind of research is missing — RCT in healthy adults, long-duration safety data, sex-stratified analyses, etc. — and whose work would close that gap.
• "Studies show…" without naming the studies. Every cited claim links to its source.
• Therapeutic outcome claims. "Treats X" / "cures Y" / "reverses Z" are banned regardless of how well-supported they seem. We describe what the literature shows, not what the molecule does in therapeutic-claim language.
• Cherry-picked supporting evidence. Where the literature has conflicting results, we present both directions with the methodological context, not the more flattering side.
• Conflict-of-interest framing without naming it. Trials sponsored by molecule manufacturers are flagged. Inventor- group trials are flagged. Compounding-pharmacy-adjacent sources are flagged. Without flattening the framing.

A meaningful share of the peptide literature is published in Russian (Khavinson group, Ashmarin group, Levitskaya group), in Croatian (Sikiric BPC-157 lineage), and in journals with constrained Western indexing. Translating this work into English is a load-bearing editorial activity for this platform — the Russian-language literature is the central reason that Selank, Semax, and Epitalon have evidence-bases the English-language peptide community is mostly unaware of.

Process for translated entries:

• Original PDF located via primary archives (peptogen.ru, selank.ru, Moscow State University biology server, St. Petersburg Institute of Bioregulation and Gerontology archive, peer-reviewed English mirrors when available).
• AI-translation pass produces a structured extraction: bibliographic metadata, abstract, methods, results, discussion. Numbers — sample sizes, doses, percentages, p-values, effect sizes — are verified verbatim against the source PDF and never paraphrased.
• Translator-confidence notes accompany every translation: which numbers were directly verified, which sections had OCR issues, what was uncertain. The notes are part of the published entry so a careful reader can audit the translation as well as the source.
• Provenance metadata (translation language, original title, translator, translation date) is in the entry frontmatter and renders on the published page. AI-translation pipeline extractions are explicitly marked as such — not labeled as human-credentialed translations.
• Companion papers in the same lineage that we have not yet translated are bibliographically recorded with their PMIDs and original-language titles, so the gap is named explicitly rather than papered over.

New peptide papers are published every week. To keep the corpus current, an automated PubMed scanner runs weekly against the corpus search terms, producing a dated queue of new papers for editorial review. The promotion decision — whether a new paper becomes a published research entry — stays editorial; the scan is automated.

The scanner queries the NCBI E-utilities API for each of the 21 peptide search terms in the corpus scope, restricted to publications added to PubMed in the last 7 days (the default). Output is a markdown queue file with bibliographic metadata and the full abstract for each paper, ready to lift into a new research entry if the editorial team judges it worth promoting.

The infrastructure is open and the script is in the repository. The scanner does not silently filter — it surfaces everything matching the search; editorial decides what's noise vs signal.

For peptides where the published literature is thin (most research-only peptides), structured aggregation of consented member self-experiments produces a citable evidence layer that no single trial covers. This is the Evidence Engine on the member platform.

Privacy invariants are load-bearing:

• Consent is opt-in per cycle. Cycles default to private. Aggregations only include cycles where the member explicitly toggled consent.
• k ≥ 5 anonymity floor. Any goal × dose × route slice with fewer than 5 consented cycles is suppressed entirely. Member identity cannot be inferred from N=1 in an aggregate.
• No free-text aggregation. Outcome notes, side-effect details, intervention notes, and check-in notes remain private to the individual cycle. Only structured fields (numeric, enum, boolean) are summarized in aggregates.
• Provenance discipline. Member-logged data (cycleSource = 'member') is never mixed with editorial-pattern data (curated from published literature) or community-self- reported aggregates (collated from public peptide-community discussions). Each source class is surfaced separately so a reader can weigh them appropriately.
• Three orthogonal data streams per consented cycle: outcome ratings + side-effect categories (from the cycle itself), paired baseline ↔ post-cycle biomarker deltas (from the biomarker surface), and structured adverse-event incidence (from the adverse-event registry). Each stream aggregates independently with the same k ≥ 5 floor.

One brand promise is load-bearing for everything else: we do not link, recommend, or grade sources of peptides.Not in any tier. Not for any peptide. Not with affiliate arrangements, not as "trusted suppliers," not as "research-grade recommendations." The encyclopedia describes molecules; it does not service the supply chain.

This rule has costs — the most common reader question we cannot answer is "where should I buy this." It also has the single biggest payoff in the brand's credibility position. Vendor-linked peptide content cannot, structurally, provide unconflicted evaluation of the molecules that vendor sells. The absence of links is the absence of conflict.

Related rule: no therapeutic-product affiliate relationships of any kind. Lab providers (Function Health, Marek, Quest) are mentioned in the context of biomarker tracking because they're how members would obtain their own bloodwork — not as paid placements. Practitioner directory listings are vetted and not paid placements.

Every editorial entry carries a lastReviewed date. Entries are updated when:

• A new primary-source paper supersedes the existing evidence base.
• The corpus monitoring queue flags a paper that changes the conclusion.
• A member or critic flags a factual error (we'd rather correct than defend; corrections are documented in the entry's history).
• The legal or regulatory landscape for a peptide changes.
• Translation provenance changes — a new translation pass improves an existing AI-extracted entry.

Cyclical content audits run quarterly. The full corpus reading list is re-ranked at each quarterly retrospective on the member platform.

Millions of adults globally use peptides whose regulatory status is gray. Some have legitimate medical indications; some are recreational or aesthetic; some occupy the long tail where "research only" labeling exists for legal cover but actual use is medical-adjacent. Whether this should be the case is a separate conversation; that it is the case is the reality the editorial scope addresses.

Better that the people doing this have the most rigorously sourced, most honest, least conflicted information about the molecules they're choosing — including the safety questions, the rebound dynamics, the cycling rationale, the upstream interventions that often matter more than the pharmacological adjunct. That's the harm-reduction position. The platform's entire structure — the four-tier system, the no-vendor-links rule, the privacy invariants, the corpus-monitoring loop, the translated-literature work — exists to make that position credible at depth.