Editorial methodology

Studies are retrieved automatically from three primary sources: PubMed (via the NCBI E-utilities API), bioRxiv (via the Crossref-hosted REST API), and ChemRxiv (via RSS). Retrieval runs on a daily schedule managed by Inngest. Every retrieved record is stored with its full metadata: DOI, PubMed ID, authors, journal, publication date, and abstract.

Deduplication is enforced at the database level. The DOI, PubMed ID, and bioRxiv ID fields each carry a unique index — a record that has already been retrieved is silently skipped on subsequent runs rather than creating a duplicate row.

We retrieve studies that mention peptides by name. The query set covers all peptides tracked on the site. The retrieval agent does not filter by outcome or conclusion before storing: a study that contradicts prevailing community belief is stored the same as one that supports it.

After retrieval, each study enters a classification pipeline. The StudyClassifier agent reads the abstract and assigns a study type from a controlled vocabulary: randomised controlled trial (RCT), cohort, case-control, case report, animal / in vivo, in vitro, review, meta-analysis, preprint, or other.

The agent also assigns a relevance score (0–1) representing its confidence that the study is meaningfully informative for the peptide(s) it mentions — as opposed to a passing citation or tangential mention. Scores below 0.30 are flagged for editor review rather than passing automatically to the published queue.

All classified studies land in the Review Queue with status pending_editor_review before any information is shown to readers. No study summary or plain- language translation is published without a human editor approving it.

The classification record — agent name, agent version, and timestamp — is stored alongside each study row, creating a durable audit trail of which model version made each classification decision.

Community signal is drawn from Reddit threads in peptide-focused subreddits. The retrieval agent pulls threads and comments, tags each item by topic (protocol discussion, vendor review, side-effect report, regulatory news, research discussion, sourcing and pricing, or general), and produces a structured sentiment summary.

A novelty score reflects how much the thread surface-forms observations not already covered in the existing literature or community archive. High- novelty items — especially side-effect reports and protocol modifications — are prioritised in the Review Queue and flagged to the editorial team.

Community signal is presented as an aggregated summary of what participants are reporting, not as editorial endorsement. Every community item that mentions a specific protocol or outcome is accompanied by the relevant study evidence rating when one exists.

All community items follow the same approval gate as studies: no community synthesis is published without human editor approval.

Vendors are scored across six dimensions, each on a 0–10 scale: certificate of analysis (CoA) quality and accessibility, pricing relative to the market, shipping reliability, customer service responsiveness, accepted payment methods, and regulatory compliance posture. The composite grade (A+ through F) is a weighted average; CoA quality and regulatory compliance carry the highest weighting because they are most directly relevant to safety.

Scores are based on mystery-shop purchases, CoA review, public regulatory records, and community feedback. Each dimension has a written rationale stored alongside the numeric score. The last mystery-shop date and last full review date are published on every vendor scorecard.

Score changes — including decreases — are recorded with a timestamp. We do not remove historical grades; the full revision history is available to editors. Vendors are not notified before a score change is published.

LegitScript status is recorded as a factual data point. A LegitScript certification does not automatically raise a score, nor does its absence lower one — it is one signal among many.

AllAboutAminos does not make therapeutic claims. This site does not state that any substance “cures,” “treats,” or “prevents” any condition. When describing what research participants or community members report, we use the language of reported observation — “reported protocols” rather than “recommended protocols,” “participants reported” rather than “studies show it works.”

Evidence ratings — supported, mixed, weak, or contradicted — reflect the quality and consistency of the available research, not a clinical determination. A rating of “supported” means the preponderance of retrieved studies are directionally consistent; it does not mean the evidence is definitive or that the substance is safe for any given individual.

Side effects and risk information appear above the fold on every peptide page. This is not configurable by content editors — it is enforced in the page template. Peptide pages that do not have a side-effects section populated are not eligible for publication.

All LLM-drafted content passes a LegalScreener step in the pre-publish gate chain before reaching an editor for review. The screener flags therapeutic claims, unauthorised health claims, and language that deviates from the claims-discipline vocabulary. Flagged drafts require explicit editor remediation — they cannot be approved in a flagged state.