Missing Scientists

What it is

Between June 2022 and February 2026, eleven U.S. scientists and officials connected to defense, aerospace, nuclear, or advanced-research programs died or disappeared. In April 2026 the FBI confirmed a holistic review of the cluster, and the House Oversight Committee opened an inquiry of its own.

This is a dossier on those eleven cases. It exists to make the sources, the gaps, and the contradictions visible so a reader can draw their own conclusions — not to hand them a verdict. Four of the eleven have publicly identified explanations. Five remain unexplained, with no public statement from law enforcement on cause or third-party involvement. Two have undisclosed causes of death. The dossier says exactly that, case by case, and stops there.

How it sources

Every source carries a tier, 1 through 8. Tier 1 is primary: law-enforcement bulletins, court filings, press-briefing transcripts. Tier 3 is local and beat reporting, close to the case. Tier 4 is national mainstream, covering a story outside its regular beat. Tier 6 is reporting that traces back to a single anonymous source and then echoes across outlets. Tier 7 — podcasts, Substack, YouTube, TikTok — is read for leads and to see what claims are circulating, and is never used as evidence. Tier 8 is foreign state-affiliated press, included where geopolitically relevant, with its orientation noted.

Every factual claim then carries a confidence rating: Confirmed, Reported, Alleged, or Speculated. A claim that appears only at Tier 4 or below gets traced back to a higher-tier origin wherever one exists, so the dossier isn't quietly resting on a chain of outlets citing each other.

The whole thing runs under a no-contact policy. No individual, family, agency, employer, or journalist is ever approached. Every word of it is built from material that was already public.

What's in it

Nine hypotheses, H1 through H9, registered before the evidence was weighed and then evaluated in both directions — evidence for and evidence against, in parallel columns — so a hypothesis can visibly fail instead of quietly disappearing. Neither the mundane nor the exotic reading gets a discount: no country or entity is pre-excluded or pre-implicated, and investigating a hypothesis is not endorsing it.

A connection diagram maps people, institutions, locations, and programs across three layers you can toggle independently: tight (documented), medium (inferred), and corkboard (speculative). The corkboard layer is off by default and labeled for what it is. A timeline runs the same material chronologically.

Contradictions and known unknowns get their own registers rather than being resolved silently — where two sources disagree, the disagreement is the artifact.

The prose is AI-generated

All of it, by Claude, under human direction. The human sets scope, sourcing standards, and neutrality rules, and reviews the drafts. Factual content traces to the public sources cited inline; the synthesis and the writing are model-produced. The dossier carries that disclosure on every page, and it belongs here too — readers should weight the writing accordingly.

Stack

The research repository is markdown and JSON: one file per case on a fixed schema, with cross-case analysis, primary-source excerpts, and an append-only research log. It's maintained through a pipeline of Claude Code prompts and versioned in git, so every revision to a claim is diffable — which is most of the point.

The site is a static build with no framework: vanilla TypeScript and CSS, D3 for the diagram and timeline, rendered from the research markdown at build time and deployed on Netlify.