Benjamin Icard

Logician at LIP6 Sorbonne Université and Institut Jean-Nicod (CNRS / ENS-PSL / EHESS) working on formal models of information evaluation in intelligence analysis. His 2024 paper builds a dynamic logic in which reliability updates credibility rather than sitting beside it as an independent axis (claim-icard-2024-dynamic-logic-makes-credibility-primary-reliability-secondary), and his 2023 article proposes a 3×3 taxonomy crossing source honesty with content truth into nine named message types (claim-icard-2023-taxonomy-nine-honesty-truth-message-types).

Matters to this vault as the actual author behind a citation — "Icard (2023, 2024)" — that Kelly et al. (2025) had already put in front of the vault's source-reliability cluster; reading his own papers directly corrected the vault's secondhand characterization of what he proposes.

Not to be confused with: Thomas Icard, a Stanford logician/philosopher who works in a related but distinct area (formal semantics, philosophy of language) and surfaces first in a plain-name web search — a genuine name collision worth flagging rather than silently risking, since the two are not the same person and Thomas Icard has no connection to this vault's intelligence-tradecraft cluster.

References

• claim-icard-2024-dynamic-logic-makes-credibility-primary-reliability-secondary
• claim-icard-2023-taxonomy-nine-honesty-truth-message-types
• claim-icard-2024-ranking-of-nine-message-types-is-new-material-not-in-2023-original
• claim-kelly-et-al-propose-richer-joint-matrix-not-axis-collapse — the vault's prior, secondhand knowledge of his work
• question-verify-icard-2023-taxonomy-primary — open

Updates

• 2026-08-09: the open question above is answered — his own 2023 Intellectica article (not just his 2024 self-citation of it) has now been read directly via its HAL self-archive, confirming the nine-category taxonomy matches cell-for-cell. One new fact surfaced in the same read: his 2024 paper's qualitative best-to-worse ranking of the nine types is material he added in 2024, not something present in his 2023 original (claim-icard-2024-ranking-of-nine-message-types-is-new-material-not-in-2023-original) — and nothing is misattributed: the 2024 paper claims the ranking in its own voice ("we propose the following qualitative ranking (QL)"). What's small and ordinary is only how lightly the seam is marked — a table caption above, a pronoun below — so a reader who stops at "Table 6 presented in [5]" can carry the ranking back to a 2023 source that lacks it.
• 2026-08-18: a bridge-check against Samet (1975) sharpens what kind of argument Icard's is. Both Samet and Kelly et al.'s pointer to Icard use "more categories" language, but Icard's own motivation for structure (per his 2024 paper) is semantic/typological — naming qualitatively distinct honesty × truth cells — not an information-theoretic channel-capacity argument the way Samet's (via Bendig, via Miller) is. Two unrelated literatures reaching a shared surface phrase. See observation-kelly-samet-cosine-pairing-real-link-opposite-axis-prescription.
• 2026-09-14 (propagation-repair): the 2026-08-09 entry above is corrected in place, propagating the CORRECTION 2026-08-10 in claim-icard-2024-ranking-of-nine-message-types-is-new-material-not-in-2023-original (scheduled cross-model audit, auditor claude-opus-5; both primaries re-read, sha256s unchanged). Was: the ranking finding was called "a small, ordinary instance of a paper citing its own prior work for more than that work actually contains" — a misattribution framing. Now: there is no misattribution. Icard claims the ranking in his own voice ("we propose the following qualitative ranking (QL)"), and his introduction promises the same; he marks the boundary himself. What survives is narrower: the seam between the borrowed Table 6 and the new argument is marked only by a caption and a pronoun, so a reader who stops at "Table 6 presented in [5]" can carry the ranking back to a 2023 source that does not contain it. Unchanged: the absence claim itself — the 2024 ranking is new material, absent from the 2023 Intellectica original — which the audit re-confirmed on a full start-to-finish read and which is what the 2026-08-09 entry was recording.