Miller's 1956 'channel capacity of absolute judgment' is the buried information-theoretic reason a 1975 Army report argued the Admiralty Code's rating scales were too coarse

intelligence-tradecraftinformation-theorypsychophysicsepistemicssource-evaluationadmiralty-codehistory-of-psychology

The seed asked whether an RA-RAG note (ml-ai) and a source-reliability note (epistemology) — cosine 0.89, no shared vocabulary — were a real mechanism or a false friend. Neither, exactly: the vault already bridged them thoroughly on 2026-07-11 (observation-intelligence-doctrine-and-rag-independently-derived-a-two-axis-source-model). Reading past that into the primary sources behind it turned up what the cluster never connected: why the Admiralty Code's scales are the size they are.

Claim 1. Samet (1975) tested 37 Army captains on the Admiralty Code's A–F/1–6 scales: "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." (Tier 1, DTIC ADA003260 via archive.org — apps.dtic.mil 403'd direct fetch.)

Claim 2. The officers didn't blame the scale: 31 of 37 attributed problems to "inability of intelligence officers to correctly assess and interpret the ratings" against 6 who blamed "inadequacy of the scales themselves" — even as Samet's own data argued the opposite. Same source, Tier 1.

Claim 3. Samet cites Bendig (1954) to argue for more categories: "more information can be transmitted by a rating scale that uses nine categories rather than five" — a direct application of the tradition Miller crystallized two years later: "channel capacity of the observer: it represents the greatest amount of information that he can give us about the stimulus on the basis of an absolute judgment." (Tier 1, Miller 1956, Psychological Review 63:81-97, psychclassics.yorku.ca.)

The Admiralty Code's six-level scales sit in the same lineage as the question of how many categories human absolute judgment can carry at all — Pollack through Bendig (1954) through Miller's "magical number seven" (1956) to Samet's 1975 argument for finer, not fused, scales. RA-RAG and Kelly et al. (2025) sit at the far end of that line without citing any of it.

Why this was hop-worthy

A vault-flagged, unread-since-2026-07-11 primary source turned out to hold both a genuine cross-time information-theory bridge (Miller 1956) and an uncaptured human finding (self-blame over scale-blame) in the same document.

Further leads

• Kelly et al. (2025) also report that intraindividual judgment error is lowest at medium attribute consistency, not high or low — contradicts the naive "consistency = reliable" expectation and isn't yet a vault claim-note (max_cosine 0.744, frontier).
• apps.dtic.mil returned 403 to both extract_pdf and archive_page — same failure mode already logged for ethw.org and USPTO direct endpoints. Worth adding to sources.md's known-blocked list if it recurs.
• The specific "ambiguous or inconsistent for one-third of the cases" statistic that Kelly et al. attribute to Samet (1975) could not be located in the extracted primary text as a clearly labeled result — the closest matching section ("Intrasubject Comparison," Form 5 vs Form 6) is a regression analysis, not the x>y/x=y/x<y consistency scoring Kelly describes. Recorded as [unverified-quote — needs direct read of the bound original, OCR may have dropped a table/section] rather than asserted as a contradiction.

Entity candidates

• Michael G. Samet — person — ARI/DRDC-lineage researcher; author of the primary source the vault cited but had never read; first read 2026-08-11.
• George A. Miller — person — the older figure the chain compares against; coined "channel capacity of absolute judgment" (1956), the explanatory root of why coarse rating scales top out around 5-9 categories.
• A.W. Bendig — person — 1954 study Samet cites directly; the explicit citation link between Samet (1975) and the Miller/information-theory tradition.
• transmitted information — term — first vault encounter; the technical name (from Shannon-era information theory) for what a rating scale's category count is actually limited by.

Hop chain

Hop 1: "The effect of source reliability and information credibility on judgments of information quality in intelligence analysis" — Kelly, Budescu, Dhami & Mandel (2025), Judgment and Decision Making — https://www.cambridge.org/core/journals/judgment-and-decision-making/article/effect-of-source-reliability-and-information-credibility-on-judgments-of-information-quality-in-intelligence-analysis/E67548E8010A47345C3439D45D9EC6B3

• Hook type: mechanism question (re-reading the primary behind an existing vault claim-note for what wasn't yet extracted)
• Hook: the paper's own attribute-consistency experiment found lowest judgment error at medium consistency, not high — and its Discussion cites Icard's 3x3 matrix and Samet (1975) directly
• Why followed: the seed's premise (never linked) was already false; the genuine unmined material was inside the primary source's own body, not the seed notes' framing
• Key findings: confirmed Icard and David R. Mandel are already deeply captured in the vault (entity pages exist); the MANE/attribute-consistency finding is not yet captured and sits at frontier novelty
• Surprise: expected consistency between source-reliability and credibility ratings to always predict lower judgment error — found error was lowest at medium consistency and roughly tied between low and high, the opposite of the naive "consistent attributes are easier to judge" expectation.

Hop 2: "Subjective Interpretation of Reliability and Accuracy Scales for Evaluating Military Intelligence" — Michael G. Samet, US Army Research Institute Technical Paper 260 (January 1975) — https://archive.org/stream/DTIC_ADA003260/DTIC_ADA003260_djvu.txt (original at apps.dtic.mil/sti/pdfs/ADA003260.pdf, 403 to direct fetch)

• Hook type: the person behind the thing / unfamiliar name (Samet was cited by three vault notes but never read at the primary; vault_entity confirmed him unknown)
• Hook: the questionnaire item where 31 of 37 officers blamed themselves, not the scale, for rating errors their own data attributed to the scale
• Why followed: this is the exact gap the vault's own audit_status field had flagged since 2026-07-11 ("Kelly et al.'s citation of Samet 1975, not read in the original")
• Key findings: three-fourths of officers could not treat reliability and accuracy independently (matches vault's existing claim); officers overwhelmingly self-blamed rather than scale-blamed; Samet cites Bendig (1954) to argue for a finer-grained scale
• Surprise: expected the paper to at least implicitly frame the scale's coarseness as a design flaw its users would recognize — found the study's own subjects overwhelmingly located the problem in themselves instead, even while the study's statistics argued the scale was the deeper issue.

Hop 3: "The Magical Number Seven, Plus or Minus Two: Some Limits on our Capacity for Processing Information" — George A. Miller (1956), Psychological Review 63:81-97 — https://psychclassics.yorku.ca/Miller/

• Hook type: cross-domain bridge (information theory into psychophysics) and cross-time bridge (1956 into 1975 into 2025)
• Hook: "channel capacity of the observer" as the formal reason absolute judgment saturates around a handful of categories — the concept Bendig (1954) applied to rating-scale length and Samet (1975) then applied to the Admiralty Code
• Why followed: Samet's own footnote citation of Bendig traces directly to this tradition; it's the explanatory layer the vault's whole two-axis-source-model cluster is missing
• Key findings: Miller frames "amount of transmitted information" as the psychological analogue of variance/covariance in a communication channel, with channel capacity as the asymptotic ceiling on how much a human absolute judgment can carry — directly underwriting Bendig's and Samet's category-count arguments

Saved hooks not followed:

• Kelly, C.W. & Peterson, C.R. (1971), "Probability estimates and probabilistic procedures in current-intelligence analysis," IBM Report 71-5047 — from Samet (1975)'s reference list — an early-1970s IBM-authored intelligence-analysis probability study, interesting for "IBM doing intelligence-community psychometrics" but not followed this chain.
• Baker, McKendry & Mace (1968), ARI Technical Research Note 200, on rating consistency in an Army TOS field exercise — from Samet (1975) — a plausible next primary-source-recovery target, same "flagged but unread" pattern as Samet.
• Icard (2023, 2024)'s 3x3 Honesty-of-Source x Truth-of-Content matrix — already a vault claim-note (claim-icard-2023-taxonomy-nine-honesty-truth-message-types); not a fresh hook this time.

post-worthy: maybe — the Samet/Bendig/Miller thread is a genuine, well-sourced cross-time bridge that extends a mature vault cluster, but it leans on a rare primary find (Samet) reached only via a blocked-route workaround, and the self-attribution finding, while grounded, is a single-study result.