Reports, white papers, and open datasets from the SignalWatch conspiracy-theory tracker, for researchers and journalists. Updated 2026-07-18. Free to use with attribution.
In-depth write-ups and methodology notes from the SignalWatch system. Free to read and share.
We clustered 807 conspiracy-positive sources by breadth, political lean, topic and platform — and eight archetypes fall out. A tiny video/social mega-spreader tier bridges dozens of theories (Alex Jones spans 31) while a long podcast tail stays narrow; political lean is predicted by topic, not loudness; pseudoscience is its own apolitical lane; and deep-state/QAnon is the connective tissue, 27% of all source-to-theory links.
Download PDF ↓
Companion to Pseudoscience on the Margin. What the demarcation literature (Hansson, Boudry, Fasce) says we got right and what we mislabeled: 'they refuse to look' isn't a leak from the conspiracy shelf, it's an immunizing strategy — the canonical feature of pseudoscience. The fix matters most for the theories that actually kill people: vaccine denial drove ~2,144 US measles cases and the Americas' loss of elimination status in November 2025.
Download PDF ↓
Wrong-about-reality claims with no hidden-villain spine — except the spine quietly reappears in the language of the claims themselves. 14 theories across four domains of science, where they live (video, podcasts, marketing tails, and now the measured crank-paper slice — the viXra harvest landed, adding 31 force-linked atoms across gravity-denial / free-energy / simulation theories). Updated with the per-theory formality profile and a forward-pointer to the Revisited method note.
Download PDF ↓
How SignalWatch turns 120 conspiracy theories into one comparable dataset: the controlled vocabularies (arena, mechanism, villain), the shared cast of actors, the anatomy of 'evidence', and what the structure reveals about why some theories spread.
Download PDF ↓
When a news outlet “matches” a conspiracy theory, is it promoting it? We stance-classified every match: the NYT and Guardian are ~97% neutral coverage; advocacy concentrates in partisan outlets. With eight years of GDELT coverage trends.
Download PDF ↓
Run the same query across seven search engines and each shows measurable blind spots and a stance: block rates span 10-100%, Bing surfaces the most pro-conspiracy content, Google the most debunks, and multi-engine search finds ~2.3x the sources.
Download PDF ↓Explorable, self-contained views of the data — hover for detail, filter in the browser. Free to open and share.
Every scored post SignalWatch has collected about the hantavirus discussion, placed by platform, colored by heat and narrative. Four emerging narratives spread thin across ten platforms — low-heat, grievance- and contempt-driven, with the violence dimension near zero. Hover any node for its readout; isolate a narrative to see how it travels.
Open interactive →Open, structured cuts of the database — spreadsheet-ready and semantic-web formats.
One row per tracked conspiracy theory (266 total). The flat, spreadsheet-ready cut of the database.
The Structured Conspiracy-Theory Ontology in RDF/Turtle (OWL). Importable into Protégé, GraphDB, or any SPARQL endpoint.
Lean is a calibrated centroid in [−1 (left), +1 (right)]. The dataset is a snapshot; figures reflect the corpus harvested to date. Questions or corrections: see the site footer.