Most news is told as separate stories. The same person, company, or institution shows up in three or four of them, and nobody draws the line. We draw the line — with sources, with a confidence score, and with an audit gate before publication.
A dossier is what we call a single investigation. Every load-bearing claim traces to a primary source — a regulatory filing, a court docket, an SEC document, a sworn testimony, or a verified public statement. No homepage URLs as citations, no LLM-only synthesis, no "according to reports."
Each source carries a tier: Tier 1 (primary documents), Tier 2 (verified secondary reporting), and contradictory (sources that disagree with our read). The contradiction count is shown on every dossier card. If we have a finding nobody else has, the proof is the source list.
We publish a numeric confidence with each file, anchored to a letter grade so you can compare across files at a glance. The scoring is conservative: we promote a claim only after independent corroboration, and we publicly mark unresolved contradictions instead of hiding them.
Anything labelled lead-only is a thread we're following — not yet a finding. We don't pretend low-confidence work is high-confidence.
Every dossier with named entities, dollar figures, or material legal exposure crosses two gates before publication: a forensic accountant who walks every dollar figure back to a primary regulatory source, and an adversarial defense attorney who attacks each claim as if a witness on cross-examination.
Files in active review are visible on the homepage under Adversarial Audit, but the underlying substance stays sealed until the gate clears. The framing is: "we know this is interesting — we're not going to publish it sloppy." Better to be late than to be sued.
Across our 10 active dossiers, 656 distinct entities — people, companies, governments, vessels, addresses — are mapped on the brain map. When the same name appears in two or more independent investigations, the brain map shows the cross-edge. Today there are 14 such cross-system entities; tomorrow more.
Hover an entity on the network → only its edges light up; everything else dims. Click → focus pins and the entity panel opens with the list of every file that contains it. That's the moat: not "we have a lot of files" but "we show you the connection between them."
The Substrate Finder intersects entity sets across all dossiers. Paste two or three names — OPENAI, ANTHROPIC, ELON MUSK — and you'll see every dossier that contains all three, plus the other entities those dossiers also share. That intersection is the substrate: the shared plumbing the news cycle covers as separate stories.
No LLM, no synthesis, no hallucination risk on this surface. The math is set intersection across the entity graph, computed locally in your browser against an 80KB index. The answer is the same every time.
The slider on the homepage scrubs from April 1 to today. Dragging it backwards dims any card whose subject-date is after the cursor, so you can see what would have been on the page on any given day. The event ticker shows the most recent newsworthy event by that date — useful for orienting yourself on a story's chronology before you read the file.
We don't run named-expert avatars — there's no "panel of experts" pretending to author each file. We don't synthesize specifics that aren't in primary sources. We don't publish anything with a fact-audit below the founder's 9+/10 bar. We don't run a paywall on completed dossiers — only the gated, pre-audit work is held back. We don't share or sell entity-search behavior, because there is no tracking on this site (check the network tab).