How briefs and confidence are produced
Our methodology is designed to make automatic coverage understandable and auditable.
Event clustering
Reports are compared only inside a 48-hour window and across different publishers. The default score combines title-token overlap, character similarity, matching numbers, and publication-time proximity.
A score is a retrieval aid, not proof that every detail is true. Publishers remain individually visible so readers can evaluate the evidence.
Coverage confidence
Confidence reflects the number of reputable independent publishers and the consistency of supported facts. It does not measure the moral importance of a story or guarantee that early reports are complete.
Brief generation
Generated copy is schema-validated, bounded in length, and instructed to paraphrase. Unsupported claims, invalid output, or a missing summarization service keep the cluster out of automatic publication.
Corrections and updates
New reports can update the source table, timeline, agreement, and uncertainty sections. Original source timestamps remain intact so changes in coverage are visible.