Collection
Structured public RSS, Atom, and official API metadata are collected on a configurable cadence. Raw snapshots are archived by day for replay and audit.
Methodology
Pagish separates source facts, synthesis, analysis, inference, and validation so readers can see what is known and what is still developing.
Clustering
Items are grouped by canonical URL, durable identifiers such as arXiv or repository IDs, normalized title signatures, trend topic, and publication timing.
Editorial Output
Original source titles are preserved internally. Public display titles and summaries are Pagish-written or source-normalized only when validation passes.
Ranking
Rankings use source authority, freshness, AI specificity, and source diversity. Composite labels are shown with components instead of unsupported precision.
Legal Use
Pagish links to sources, stores compact metadata and abstracts where allowed, and does not republish full copyrighted article bodies or publisher article images.
Limits
Single-source clusters are labeled as source-limited. Company benchmark claims are not treated as independently validated unless a reproduction source is attached.