Methodology & Editorial Standards

AI Role — Narrow and Auditable

Proprietary models are used exclusively for scale tasks: bulk document ingestion, entity extraction, timeline construction, and initial relevance scoring against our coverage ontology. Every AI-generated extraction is logged with model version, prompt hash, and confidence score. No sentence that appears in a published story is written by a model.

Human Verification Layer

Every claim is dual-sourced or directly attributed to a primary document. Two editors (one of whom did not use the AI tooling for that story) must sign off before publication. Hallucination or fabrication is treated as a firing offense; we maintain a zero-tolerance internal policy with public correction notices.

Source Philosophy

We prefer primary public records obtained via FOIA, litigation, or direct agency disclosure over secondary reporting. When we use whistleblower material, it is corroborated against documentary evidence wherever possible and labeled clearly when it cannot be.

Ethics & Conflicts

We do not accept funding or access from the institutions we cover. We maintain a strict gifts-and-hospitality policy. Staff and advisors recuse from any story in which they have a material personal or financial interest.

Full methodology appendix available to institutional subscribers. Corrections policy published on every story page.

© 2026 CLOSED PRESS