Evidence and limits

The argument stays inspectable

Observed facts, inferences, stress-test assumptions, and policy proposals are different kinds of claim.

  • 25 source notes
  • 29 HTTPS links
  • Versioned for public review

How to read the claims

This project separates four kinds of statement:

The site does not convert an inference into a fact by repeating it more confidently.

Important limits

Publication record

This public working draft is intentionally versioned for iteration. The source manuscript passed 39 builder checks and 29 independent adversarial checks before the minisite build. The publication build adds route, link, fragment, download-hash, accessibility-structure, responsive-layout, visual, and production-integrity checks.

The current paper contains 25 source notes and 29 HTTPS source links. Direct server checks can receive automated-access denials from some government sites; the research corpus preserves the captured source material used for review.

Source ledger

The notes below are the paper's complete public source list. Each note retains the scope or limitation stated in the manuscript.

06

National Institute of Standards and Technology, IR 7946, CVSS Implementation Guidance, §2.2, 2014. This document addresses CVSS v2 and is cited for its explicit discussion of contextual risk and chaining limitations, not as the current CVSS specification.

08

Defense Advanced Research Projects Agency, “AI Cyber Challenge Results”, Aug. 8, 2025. DARPA corrected the synthetic-vulnerability denominator from 70 to 63.

23

Dario Amodei, The Adolescence of Technology, Jan. 2026, §2 (“Ability and motive may even be negatively correlated” and “this will break the correlation between ability and motive”). The paper applies Amodei's general destructive-misuse mechanism to cyber; it does not attribute Jimmy Staley's paraphrase to Amodei as a quotation.