How we use automation

Automated checks

Before publication, automated checks look for missing citations, invalid dates or status labels, duplicate records, and personal or administrative information that does not belong on a public page.

The narrow automated lane

Selected ClinicalTrials.gov records and retained PubMed metadata may enter bounded Tier A lanes limited to structured source facts. Each candidate must pass deterministic eligibility checks and one isolated AI review against the publication contract. Ambiguous records and anything containing dosing, sourcing, vendor promotion, individualized advice, or unsupported claims stay unpublished.

The exact trial-publication graph is narrower still: deterministic software reads only structured NCT and PMID fields for records already in the reviewed public authorities. It does not use an AI model, titles, abstracts, authors, or peptide-name similarity to create an edge.

Human governance

People approve source classes and release boundaries, confirm identity mappings, review interpretive summaries, and handle reports and corrections. The broader discovery queue does not inherit either narrow Tier A lane, and the PubMed lane does not summarize article content.

What does not publish through this lane

The lane does not cover social posts, secondary reporting, vendor material, medical recommendations, or free-form claims about safety or effectiveness. Coverage remains selective, and new source activity may not appear immediately.

Use of AI

AI may classify whether a structured candidate fits the narrow contract and may assist with research, extraction, and editing. It does not establish medical truth. Every public factual statement remains attributable to a named source, and deterministic gates can still hold or reject the record.