METHODOLOGY
How Optics reads a headline cluster.
Optics is a headline-framing detector. It compares how outlets word the same event, then surfaces source spread, loaded words, content format, and confidence.
Optics scans 380+ RSS feeds from 250+ outlets every 5 minutes.
Articles are clustered by title overlap, entities, bigrams, description text, and sanity checks.
Multi-source stories get a 0-99 headline-contrast score plus loaded-word highlights.
High requires 6+ sources and 3+ bias buckets. Medium requires 3+ sources and 2+ buckets. Two-source or heuristic matches stay developing.
Optics flags news-vs-analysis, think-tank, opinion, podcast, live-update, review, and deal formats so high gaps are not overread.
Bad matches, source-label disputes, and misleading Wording Gaps can be sent to corrections@optics.news for review.
How to read the score
- 0-20: similar framing.
- 21-50: mild wording shift.
- 51-75: strong wording shift.
- 76-99: major first-impression split.
Known limits
- Headline and RSS-description analysis only.
- RSS-only source ingestion.
- Low-confidence two-source clusters can reflect format differences as well as framing.
- Bias labels are outlet-level context, not article-level truth claims.
Examples of match quality
- High confidence: 6+ sources, all three major buckets, same article format.
- Medium confidence: 3-5 sources, at least two source buckets, no major format mismatch.
- Developing: two sources, missing center coverage, or news compared with analysis/opinion.
- Early pickup: one to three sources, useful to watch but not enough to call a blindspot.
Source labels
Bias and factuality labels are outlet-level context. MBFC is the baseline label source while AllSides and Ad Fontes disagreements are tracked where available. Optics does not independently assign whole-outlet ideology scores.