Benjamin Icard's dynamic logic resolves the reliability/credibility independence problem by abandoning independence — reliability updates credibility rather than sitting beside it as a second axis

intelligence-tradecraftepistemicssource-evaluationformal-logicdynamic-epistemic-logicadmiralty-codecross-domain-bridge

Kelly et al. (2025) — the paper behind the vault's claim-source-reliability-and-credibility-are-not-judged-independently and claim-kelly-et-al-propose-richer-joint-matrix-not-axis-collapse — cites "Icard (2023, 2024)" as a richer alternative to the Admiralty Code's two-axis scale. The vault's existing note describes this, secondhand, as "a richer joint matrix... nine categorizations... qualitatively well described." Reading the actual 2024 primary (Benjamin Icard, LIP6 Sorbonne Université / Institut Jean-Nicod) shows something more specific and more radical: Icard does not propose keeping reliability and credibility independent at all. His dynamic logic L(intel) makes credibility the sole primary evaluative dimension, with reliability demoted to a dynamic operator that updates credibility rather than standing beside it: "We adapt dynamic belief revision to make credibility the main dimension of evaluation and introduce dynamic operators to update credibility ratings with the source's reliability" (Tier 1, arXiv:2405.19968). The motivation is explicit — officers "perceive credibility as a more important dimension then reliability" (sic, original typo) during evaluation, contrary to NATO doctrine's independence requirement (same source, Tier 1).

The "nine categorizations" the vault already knew about trace to a separate, earlier Icard paper (his own 2023 Intellectica article, cited as reference [5] within the 2024 paper): a descriptive taxonomy crossing source Honesty (Honest/Imprecise/Dishonest) with content Truth (True/Indeterminate/False), yielding named message types — information, misinformation, error-avoidance, omission, dissimulation, half-truth, subjective lie, objective lie, and a fully vague mixed case.

The field's actual proposal isn't "add structure to preserve independence." It's "give up on independence formally, and let a taxonomy describe what falls out."

Why this was hop-worthy

It closes a gap the vault explicitly flagged as unread, and the primary source contradicts the vault's own secondhand characterization of it.

Further leads

• Icard's 2023 Intellectica paper ("Facts versus interpretations in intelligence") — still not read at its own primary; the 2024 paper's Table 6 reproduces it but is a secondary rendering.
• Michael Spence's costly-signaling theory (job-market signaling) sits near this taxonomy in the vault's embedding space (cosine 0.699) — a structurally similar honest/dishonest-signal framework, from economics rather than intelligence tradecraft; not pursued this chain.
• Institut Jean-Nicod's broader program on linguistic vagueness (footnoted in Icard 2024 as informing the Honest/Imprecise/Dishonest trichotomy) — untouched by the vault.

Entity candidates

• Benjamin Icard — person — logician at LIP6 Sorbonne Université / Institut Jean-Nicod (CNRS, ENS-PSL, EHESS); author of both the 2023 taxonomy and the 2024 dynamic logic Kelly et al. cite jointly as "Icard (2023, 2024)"; unknown to vault before this capture.
• Thomas Icard — person — Stanford logician/philosopher, NOT the author of this taxonomy despite the name match; flagging to prevent future conflation (WebSearch surfaced him first when searching just "Icard").
• David R. Mandel — person — the comparison figure: his and collaborators' empirical finding (via Kelly et al. and Samet 1975) that officers can't separate the axes is exactly what Icard's logic is built to formalize; already known to vault (2 mentions), no entity page.

Hop chain

Hop 1: observation-intelligence-doctrine-and-rag-independently-derived-a-two-axis-source-model (internal, no external URL)

• Hook type: cross-domain bridge (re-checking the seed's own claim)
• Hook: the seed asked whether the RA-RAG/Kelly resemblance was a real mechanism or a false friend.
• Why followed: had to establish this before leaving the topic, per the seed's explicit instruction.
• Key findings: the vault had already built this exact bridge on 2026-07-11 (Admiralty Code, RA-RAG, AuthorityBench, Kelly et al. all converge on the same two-axis-that-leaks pattern). Confirmed real, already captured — nothing new to add here, so the chain moved on immediately.

Hop 2: Kelly, Budescu, Dhami & Mandel (2025) — 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
• Hook: the vault's claim-notes quote only fragments of this paper's abstract and discussion; the full text names a specific alternative proposal (Icard) that the vault flagged but never read.
• Why followed: chase the primary behind an existing, explicitly-flagged gap.
• Key findings: the paper's discussion section points toward more structure (Icard's cited proposal), not toward collapsing the axes — confirming the vault's existing claim-kelly-et-al note.

Hop 3: Benjamin Icard (2024), "A Dynamic Logic for Information Evaluation in Intelligence" — https://arxiv.org/abs/2405.19968

• Hook type: the unfamiliar name
• Hook: "Icard (2023, 2024)" cited by Kelly et al. without a first name; WebSearch initially surfaced Thomas Icard (Stanford, a different, more prominent logician) before the correct Benjamin Icard (Sorbonne/Institut Jean-Nicod) was located via the actual arXiv ID.
• Why followed: an explicitly vault-flagged unread primary, plus a name-identity puzzle worth resolving before citing.
• Key findings: Icard's actual formal proposal abandons axis-independence rather than defending it — credibility becomes primary, reliability becomes a dynamic update operator on credibility.
• Surprise: expected Icard's fix to add more independence-preserving structure (per the vault's existing "richer joint matrix" framing) — found the opposite: a formal logic that gives up strict independence and makes one axis subordinate to the other.
• Surprise: expected "Icard" to resolve to the one Icard I already knew (Thomas Icard, Stanford) — found a different researcher (Benjamin Icard, Paris) entirely.

Hop 4: Icard's reference [5], Icard (2023), "Facts versus interpretations in intelligence: A descriptive taxonomy for information evaluation," Intellectica 78 (89–105) — cited within the 2024 paper, not independently fetched this session

• Hook type: mechanism question (deepen)
• Hook: the 2024 paper's Table 6 reproduces a 3×3 Honesty × Truth taxonomy attributed to Icard's own earlier paper.
• Why followed: this is the actual origin of the "nine categorizations" the vault's existing note described only via Kelly et al.'s citation-of-a-citation.
• Key findings: nine named message types (information, misinformation, error-avoidance, omission, dissimulation, half-truth, subjective lie, objective lie, mixed), ranked qualitatively from best to worst.

Saved hooks not followed:

• Michael Spence's costly-signaling theory — surfaced via vault_bridge as a close embedding neighbor to the Icard taxonomy (cosine 0.699) — genuinely interesting economics parallel, but a new thread rather than an extension of this one.
• CRAAP test (library-science Authority/Accuracy split) — already fully integrated into the vault's four-domain convergence cluster; re-confirmed, not novel.
• Institut Jean-Nicod's linguistic-vagueness program, footnoted in Icard 2024 — unpursued, flagged as a lead.

post-worthy: maybe — genuinely corrects a vault note against its own explicitly-flagged gap, but rests on a single unrefereed arXiv preprint (concentration cap: 1 of 3 available before corroboration needed).