Trust layer
Methodology
How DeadRisk reads the record — and the lines we will not cross.
What we do
DeadRisk turns twelve years of tech-news coverage into instruments a trader can use. The substrate is TEXXR: about 132,000 articles since 2014, roughly 256,000 extracted relationships between companies, people, products, and technologies, and vector embeddings that let us measure how a story shifts rather than just count headlines.
From that base we build three kinds of reading:
- Momentum. How fast coverage of a name is rising or falling, measured against its own trailing baseline — not against a vibe.
- Narrative shift. How the framing of a name changes over time — safety, enterprise, developer, consumer — read from the embeddings, not from keywords.
- Second-order. Which announcements touched a company's world before the market wired the connection, traced through the relationship graph.
The library
The library is our reference shelf: serious nonfiction — finance, economics, behavioural science — indexed so we can quote a framework from the source instead of reciting it from memory. When a thinker's lens shapes an analysis, the lens page and the book's own page cite the passage it rests on. That is how the Graham, Shiller and Thaler lenses stay honest: applied to what is happening now, grounded in the text, and checkable — every book on that shelf is one you can go and read.
The rules
These are enforced on every page, not aspired to:
- Coverage intelligence, not investment advice. We report what the record shows. We never tell anyone what to buy or sell.
- No price targets. No buy, sell, or hold language. No return promises. No portfolio advice. "What to Watch" lists signals in the coverage record — observable events — not trade ideas.
- Every number has a baseline. A figure with nothing to compare it to is not published. "Compared to what?" is the first question of every chart.
- Every claim traces to a source. Article IDs, edge IDs, and cited passages back the factual claims. Claims about living people come only from the cited coverage.
- At least one corpus-only fact per page. A number, edge, or timeline only our data could produce. If a reader can get it from a headline feed, we do not publish it.
The editor
Greg Salwitz is the named editor of everything here. Research desks and drafting tools operate under the editor; nothing runs pseudonymous-only. If a page is wrong, that is an editor's error, and the correction is the editor's job.