How we earn the grade.

Every report combines structured methods: Indications & Warning (I&W) monitoring, freeform OSINT collection, and, on deeper work, Analysis of Competing Hypotheses (ACH). The point is repeatability — a second analyst working the same evidence should reach the same judgment.

Findings are stated bottom-line-up-front, then supported. We report what the evidence carries and no more.

Every source carries a two-character code: a reliability letter (the source's track record) and a credibility number (whether this specific item is corroborated). An A1 is a reliable source with confirmed information; a C3 is fairly reliable and possibly true. You always see how much weight the evidence holds.

Estimative words carry fixed probability bands, never rhetorical emphasis. When a report says almost certainly, it means 95–99% — not a figure of speech. This is what lets you act on a judgment without re-interpreting it.

Collected artifacts are hashed (SHA-256) and logged under a Berkeley-Protocol chain of custody. A hash manifest ships in every report, so any reader can verify the integrity of the underlying material independently. An artifact that could not be retained or hashed is itself logged.

Before delivery, every report passes an independent red-team challenge that tries to break its own thesis. It is not an add-on and not a tier upgrade — it runs on every report, always.

Where the challenge weakens a judgment, the judgment is downgraded and the change is shown. The goal is to fail the thesis here, before you rely on it.

Between what the Corvus engine asserts and what the record treats as true, a human analyst signs off. Autonomy accelerates collection and structuring; it does not replace judgment. Nothing reaches your library unreviewed.