In Samet's 1975 Admiralty Code test, three-fourths of officers treated the reliability rating as correlated with accuracy rather than as an independent dimension

intelligence-tradecraftsource-evaluationadmiralty-codecognitive-biasepistemics

Michael G. Samet's 1975 US Army Research Institute study (ARI Technical Paper 260) tested 37 Army captains on the Admiralty Code's A–F reliability and 1–6 accuracy/credibility scales and found that most could not use the two axes as the scheme's design requires. In Samet's words: "Approximately one-fourth of the subjects treated reliability and accuracy as independent dimensions; the other three-fourths of the subjects treated the reliability rating as highly correlated with the accuracy rating."

This is the primary-source grounding for a finding the vault had carried second-hand. The claim that evaluators cannot keep source reliability and information credibility independent rested on Kelly et al. (2025) and on Kelly's citation of Samet, with the original unread since 2026-07-11. Reading Samet directly supplies the number at the source: three-fourths of trained officers collapsed the two axes into one correlated judgment. The same non-independence, now known by name, is an instance of the halo effect — a global impression bleeding across ratings that are supposed to be judged apart.

The finding indicts the two-axis design's central assumption from inside its own user population, and it sits at the head of a long recurrence: the same reliability-bleeds-into-credibility failure resurfaces in 2025 retrieval-augmented generation, where a document's own text degrades an LLM's judgment of its source's authority. Note the scope: this is the reliability-vs-accuracy correlation finding, distinct from the separate "one-third of cases ambiguous or inconsistent" figure Kelly attributes to Samet, which a direct read could not locate (see question-verify-samet-1975-one-third-inconsistent-figure).