The Graph Editors Build
Quality follows the curated source graph humans maintain—not prompt cleverness or whatever links are easiest to fetch.
Papyrus treats references as first-class editorial artifacts. Reliability depends on what evidence enters the curated structured knowledge graph and how humans evaluate it.
Reviewers assess relevance, reliability, and topical fit. Ratings and comments become steering memory for future research and planning. They are not votes for engagement—they are judgments about what the publication is willing to reuse.
This prevents drift toward sources that are merely easy for agents to retrieve. Editorial standards stay explicit.
Over time, curated reference history becomes institutional memory: what the newsroom knows, why it considers that knowledge credible, and how sources relate to one another.
That memory lets later editions explain trends instead of summarizing isolated links. New material connects to accepted context the editors have already defended.
Models are neither neutral nor magically trustworthy. The publication's discipline is human governance of evidence—accept, reject, annotate, and revisit as mission or facts change.