Discover
Start from focused recipe, medicine-making and method collections. Only links that match the source profile are admitted to the queue.
The importer is deliberately split into discovery, research, drafting and publication. A successful crawl creates traceable research topics — never an automatically published copy.
Keeping these layers separate prevents noisy archive pages, duplicate recipes and unreviewed health claims from leaking into the public journal.
Start from focused recipe, medicine-making and method collections. Only links that match the source profile are admitted to the queue.
Collections, articles, gated pages, promotions and irrelevant commercial pages are scored differently. Category pages discover links but never become editorial topics.
Canonical, title, exact-content, recipe and MinHash signatures are calculated. Metadata-only sources do not retain copied full text.
Exact and near-identical pages attach to one research topic while preserving every source URL, publisher and similarity decision.
An optional local model creates an original structured draft from the strongest source trails. The output is always saved as needs_review.
A named reviewer checks recipe fidelity, evidence wording, interactions, red flags and source citations before approval or publication.
This distinction is the reason a healthy crawl can produce topics while the public article count remains zero.
Each source has its own collection paths, article patterns, exclusion terms, minimum content threshold and article signals.
Recipe publishers can establish preparation history and technique. Clinical or regulatory claims require a separate authority layer, explicit citation and qualified review.