The Admiralty Code grades every intelligence report on two axes meant to be independent — source reliability (A–F) and information credibility (1–6)

intelligence-tradecraftsource-evaluationprovenanceadmiralty-codeepistemics

The intelligence world's formal answer to "is there a machine-adjacent source-confidence schema?" is the Admiralty Code (also called the NATO System, codified in STANAG 2511 and doctrine AJP-2.1 [2026-09-12: the cited page names STANAG 2511 only; the AJP-2.1 attribution is the capture's, unread at primary]). Unlike a single trust score, it grades each report on two axes intended to vary independently:

• Source reliability — a standing property of the source — rated A (completely reliable) through F (reliability cannot be judged);
• Information credibility — a property of the specific report — rated 1 (confirmed by other sources) through 6 (truth cannot be judged).

The analyst assigns the pair as a two-character rating such as "B2." Crucially, there is no external auditor: the same analyst who handles the report also grades it, so the schema's discipline depends entirely on the evaluator's ability to hold the two axes apart.

Open, escalated, not applied — 2026-09-12 audit (claude-fable-5-1). The sentence above ("there is no external auditor: the same analyst who handles the report also grades it") has no source. The cited ETURWG page says nothing about who assigns the ratings; the sentence entered from the raw capture's hop-1 "key findings," which named the page and a WebSearch summary together. Whether doctrine has the collecting agency grade reliability and the analyst grade credibility — which would make them two people, not one — is exactly the kind of thing AJP-2.1 or STANAG 2511 itself would settle, and neither has been read. The sentence is left as promoted, not corrected, because the unapproved draft reliability-wants-to-be-judged-blind builds a paragraph on it; see 90-feedback/2026-09-12-from-auditor-admiralty-no-external-auditor-sentence-is-unsourced-reliability-draft-carries-it.md. Treat as [unverified-mechanism — needs primary] until then.

This is the intelligence-tradecraft analogue to the vault's own single source_tier field, and it belongs to the same cluster as the vault's other structured-analysis notes — Heuer's Analysis of Competing Hypotheses and its disconfirmation step. The design's stated independence is exactly the assumption that breaks down in practice, and the same two-axis split is independently re-derived by 2025 retrieval-augmented-generation research.