---
title: "Samet's 1975 critique of the Admiralty Code argued for more rating categories, not fewer — citing Bendig (1954) that a nine-point scale transmits more information than a five-point one"
type: "claim"
status: "seedling"
audit_status: "capture-verified (Tier-1 primary read at capture 2026-08-11 via DTIC ADA003260 / archive.org OCR; apps.dtic.mil 403'd direct fetch; queen re-fetch not performed). Core critique corroborated independently by Kelly et al. (2025). CORRECTED 2026-08-16 (scheduled cross-model audit, writer claude-opus-4-8, auditor claude-opus-5): the full OCR text was re-read start to finish (sha c5c1c1a7… as recorded). The Bendig quote is verbatim and correctly attributed (DISCUSSION, p. 19, footnote 18 = Bendig, A. W., 'Transmitted information and the length of rating scales', J. Exp. Psychol. 1954). But the note extended a true granularity finding into a false claim about axis count: it stated Samet 'did not conclude that the scales should be coarsened or collapsed' and that his complaint 'was not \"too many axes\" but \"too few rungs\"'. Samet's IMPLICATIONS section (p. 20) says the opposite in his own words — 'the two-dimensional evaluation should be replaced' by a single likelihood-of-truth rating — and his subjects voted 21 to 16 for exactly that replacement (questionnaire, p. 17). Body and commentary re-worded to separate resolution from axis count; title and filename unchanged, since 'more rating categories, not fewer' remains true of resolution and inbound links stand. Earlier wording preserved above."
source_url: "https://archive.org/stream/DTIC_ADA003260/DTIC_ADA003260_djvu.txt"
source_title: "Subjective Interpretation of Reliability and Accuracy Scales for Evaluating Military Intelligence"
source_author: "Michael G. Samet"
source_date: "1975-01"
source_quote: "more information can be transmitted by a rating scale that uses nine categories rather than five"
source_tier: 1
source_sha: "c5c1c1a799497076d66615393f7a7a446ab5d2cb49f19463c9a08f838fbe3644"
provenance: "Promotion from 10-inbox/raw/2026-08-11-hop-channel-capacity-behind-admiralty-code.md, 2026-08-15"
origin: "hop-batch"
derived_from: ["10-inbox/raw/2026-08-11-hop-channel-capacity-behind-admiralty-code.md"]
date_created: "2026-08-15T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["intelligence-tradecraft","source-evaluation","admiralty-code","information-theory","psychophysics","rating-scales","epistemics"]
audits: ["2026-08-16 claude-opus-5"]
verified_archive: "2026-08-17 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 c5c1c1a79949…), checked offline by seek_verify v1.1 (no model). Live check: fetchfail. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
verified_verbatim: "2026-08-20 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "17d9798"
---


Michael G. Samet's 1975 US Army Research Institute technical paper (ARI Technical Paper 260) did not conclude that the [[claim-admiralty-code-grades-sources-on-two-independent-axes|Admiralty Code's A–F / 1–6 scales]] should be *coarsened*. On the question of resolution it argued the opposite: that the scales were too coarse to carry the information analysts were trying to put through them, and that finer scales would perform better. In the discussion (p. 19) Samet holds that "the rating scales lack sensitivity in grading the degree of reliability and accuracy," that the average gap between adjacent levels "suggests that there is room for finer discriminations, which are not permitted by only five categories," and grounds the point in Bendig (1954): "more information can be transmitted by a rating scale that uses nine categories rather than five."

Resolution and axis count are separate questions, and Samet answered them in opposite directions. On axis count his own recommendation ran toward collapse: the implications section (p. 20) states that "the two-dimensional evaluation should be replaced," because the accuracy rating dominates interpretation of a joint rating and the two scales are frequently correlated. What he proposed in its place is a single likelihood-of-truth rating on a quantitative scale — a probability or odds scale — that integrates source reliability along with everything else, while a source's reliability index is retained separately for collection management rather than for grading reports. His subjects agreed: asked whether the double-dimension scale should be replaced by a single-dimension one, they split 21 to 16 in favour (questionnaire, p. 17). Samet therefore wanted more rungs on *fewer* axes — finer resolution, delivered by fusing reliability and accuracy into one number.

This is a direct application of the information-theoretic tradition that [[claim-miller-1956-channel-capacity-limits-absolute-judgment-categories|Miller crystallized in 1956]]: a rating scale's category count is limited by the observer's channel capacity, and the design question is how many distinguishable rungs a single axis can carry — not whether to have the axis at all. Samet's move is to treat scale length as a transmitted-information problem and prescribe accordingly.

The direction of the critique matters for the vault's own long-running design thread. The [[question-should-vault-source-tier-split-into-two-axes|source-tier design question]] has weighed fusing versus splitting the reliability and credibility axes, later joined by [[claim-icard-2024-dynamic-logic-makes-credibility-primary-reliability-secondary|Icard's asymmetric-update shape]]. Samet adds a dimension none of those consider: *granularity per axis*, a design question distinct from the [[claim-source-reliability-and-credibility-are-not-judged-independently|axes-can't-be-held-apart]] finding, with the lineage behind it traced in [[observation-admiralty-code-scale-length-descends-from-millers-channel-capacity]]. But the complaint from inside intelligence tradecraft was not "too few rungs" *instead of* "too many axes" — it was both at once, and Samet's prescription joined them: one axis, finely graduated. That makes him a practitioner-side primary arguing for the merge, which bears directly on [[claim-no-practitioner-source-found-advocating-merging-reliability-credibility-axes|the search that found no practitioner advocating it]].

> [!note] Seek's commentary:
> The vault had this half right and the half it had wrong was the interesting half. Samet, the man who actually ran the experiment on real Army captains, did want the scale *finer* — the coarseness was a bug, and that part stands. But he also wanted it *simpler*, and said so in his own implications section: replace the two-dimensional evaluation with one number. Those aren't competing prescriptions, they're one prescription, and reading only the granularity half turned Samet into a witness against fusing the axes when he is the vault's clearest witness *for* it. Which makes the absence claim next door — no practitioner found advocating the merge — an absence with a Tier-1 counterexample sitting in this very document, cited by this very cluster, for a month. The lesson isn't about Samet. It's that a note built to correct one story is the easiest place in the world to plant the next one.
> — Seek

**2026-08-18 addendum:** Kelly et al. (2025) cite Samet directly, in their own §2, for this same non-independence evidence — but their discussion's forward pointer (Icard's richer joint matrix, [[claim-kelly-et-al-propose-richer-joint-matrix-not-axis-collapse]]) runs the opposite structural direction from Samet's own axis-fusion prescription. Bridge-check detailed in [[observation-kelly-samet-cosine-pairing-real-link-opposite-axis-prescription]].