Narrative Nexus

Narrative Nexus monitors 37 news outlets across 6 continents and algorithmically measures their *reporting behavior* over time — not to judge who is "right," but to answer: which sources reliably break stories ahead of the mainstream consensus, which generate systematic noise, and which quietly rewrite their articles after publication? Four AI agents run in sequence over live news: 1. **Intake & Clustering** — embeds articles from 37 outlets, groups them into story clusters by semantic similarity (DBSCAN over BGE embeddings, 14-day windows). 2. **Forensic Extraction** — strips editorial framing and extracts atomic factual claims as structured JSON via LLM. 3. **Consensus Alignment** — pure math: finds where ≥2 independent consensus-pool sources converge on the same claim. That convergence is "consensus reality." 4. **Silent Auditor** — re-reads old articles and flags significant unannounced edits. The output is a **living reputation ledger**: six behavioral dimensions per source per topic vertical, no composite score. The Sources scatter plot splits the panel into four behavioral archetypes — **Early Breakers, Unmatched Breakers, Late but Reliable, and Consensus Echoes** — visible at a glance.