docs(taxonomy): #498 AIF chantier PR-2 — Faulty comparison cluster (proposition, gated) by jsboige · Pull Request #701 · ArgumentumGames/Argumentum

jsboige · 2026-07-05T09:08:25Z

Summary

Second cluster of the #498 chantier (continues PR-1 #699 — False analogy sub-sub). Models the Faulty comparison sub-sub (Soussousfamille=Faulty comparison, pk 834-838) of the Fallacious comparison subfamily. Proposition only — GATED, 0 write to prod CSV.

Cluster shape (differs from PR-1)

Unlike PR-1 (which had an in-sub-sub mapped anchor, pk 839), this cluster has no in-sub-sub mapped anchor → it borrows the d2 root pk 833 ("Comparaison fallacieuse", MAPPED). 833 uses a direct-conflict pattern (DirectRef to BiasedClassification_Conflict + ExceptionSimilarityCase_Conflict, ExceptionRef empty), unlike PR-1's anchor 839 (exception pattern). Leaves follow whichever pattern honestly fits.

5 leaves modeled along a clean semantic fault line

836 Classification non exclusive + 838 Distinction sans différence → fully-modeled, honest scheme-divergence (classification leaves in a comparison sub-sub): VerbalClassification_Inference / ArbitraryVerbalClassification_Inference + BiasedClassification_Conflict / PropertyNotExistant_Conflict.
834/835/837 (abusive/incomplete/incoherent comparison) → FAIL-LOUD on the scheme: the 36-scheme AIF vocabulary has no native Comparison_Inference (verified), so they take a DirectRef (ExceptionSimilarityCase_Conflict, mirroring anchor 833) but document the absent scheme in AIF_skosOther rather than fabricating one.

jsboige enriched-DoD compliance (§7)

Per jsboige's last comment on #498, AIF modeling decomposes the argument into I-nodes (premises/conclusion) + RA-node (scheme) + CA-node (conflict, typed undermine/undercut/rebut). §7 maps the 2 fully-modeled leaves:

836 = undercut on the VerbalClassification_Inference RA-node (categories overlap → inference defective).
838 = undermine on the premise (distinguishing property does not obtain).
834/835/837 = FAIL-LOUD at the RA-node level (no RA to attack — a deeper gap than PR-1's 840 "no CQ").

The I/RA/CA decomposition is recorded in this doc but not yet serialized — it would require new CSV columns (AIF_attackType, AIF_attackedNode) for the terminal target. Ratification sought before any CSV schema evolution.

Native-vocabulary discipline

All 9 proposed AIF tokens confirmed in existing usage (26 Conflict / 36 Inference schemes across the 70 mapped rows). Comparison_Inference appears only in the negation (absent), never as a proposed token. Validated programmatically. No fabrication.

Honest coverage accounting

DoD-honest (legitimate scheme + CQ): 74 → 76 fully-modeled.
DirectRef-loose (CSV's looser criterion): 74 → 79.
Both reported; 76 is the feat(ontology): modéliser les sophismes comme exceptions à des schemes (paradigme AIF/Walton), au-delà des ~20 natifs #498-relevant number.

Gate (HARD)

❌ No prod CSV write, no DB write, no OWL regen, no aif: assertion emitted.
❌ No [runtime pending] feat(dnn): #649 RulesExplorer i18n — mirror loc() culture cascade #674/chore(dnn): #131 migrate 12 RazorComponent templates to Razor14 (source-level, runtime pending) #596 merge, no régén launch.
✅ Proposition only, derived code=truth from the CSV + native AIF vocabulary.
✅ Honest scheme-divergence on 836/838; FAIL-LOUD on the scheme for the 3 pure-comparison leaves (DirectRef preserved).

Relates

dispatch yj7u3j (tertiary/idle), #498 (chantier), PR-1 #699 (False analogy, predecessor), #133/#130 (OWL), #499 (inverse: virtue = good tenor), #192, #458.

— po-2023

…roposition, gated) Second cluster of the #498 chantier. Models the Faulty comparison sub-sub (pk 834-838) of the Fallacious comparison subfamily. Proposition only — GATED, 0 prod CSV write. Cluster shape (differs from PR-1): no in-sub-sub mapped anchor → borrows the d2 root pk 833 ("Comparaison fallacieuse", MAPPED). 833 uses a direct-conflict pattern (DirectRef to BiasedClassification_Conflict + ExceptionSimilarityCase_Conflict, ExceptionRef empty), unlike PR-1's anchor 839 (exception pattern). The leaves follow whichever pattern honestly fits. 5 leaves modeled along a clean semantic fault line: - 836 Classification non exclusive + 838 Distinction sans différence → fully-modeled, honest scheme-divergence (classification leaves in a comparison sub-sub): VerbalClassification_Inference / ArbitraryVerbalClassification_Inference + BiasedClassification_Conflict / PropertyNotExistant_Conflict. - 834/835/837 (abusive/incomplete/incoherent comparison) → FAIL-LOUD on the scheme: the 36-scheme AIF vocabulary has no native Comparison_Inference (verified), so they take a DirectRef (ExceptionSimilarityCase_Conflict, mirroring anchor 833) but document the absent scheme in AIF_skosOther rather than fabricating one. Native-vocabulary discipline: all 9 proposed AIF tokens confirmed in existing usage (26 Conflict / 36 Inference schemes across the 70 mapped rows). Comparison_Inference appears only in the negation (absent), never as a proposed token. Validated programmatically. pk 838 moves here from PR-1's earlier draft (boundary correction #699 — 838 is in Faulty comparison, not False analogy). Honest accounting: 74→76 fully-modeled (DoD-honest) / 79 DirectRef-loose, both reported. Co-Authored-By: Claude-Code <noreply@anthropic.com>

…/CA-node) for chantier parity with PR-2 PR-2 #701 carries a §7 mapping each leaf to jsboige's enriched DoD (decompose into I-nodes + RA-node + CA-node, typed undermine/undercut/rebut). PR-1 was missing the parallel section — inconsistent for an ai-01 batch review that will read both PRs together. Added §7 to PR-1: maps the 4 False-analogy leaves to the decomposition. - 841 Analogie étendue → undercut (similarity not transitive; inference defective). - 842 similarité fallacieuse → undercut (resemblance too superficial). - 843 Fausse équivalence → undermine (premise "X ≡ Y" is false). - 840 Pétition analogique → FAIL-LOUD at CA-node level (RA exists, CA missing — circularity has no native AIF Conflict node). Attack-type distribution: 3 of 4 are undercuts (consistent with jsboige's "most fallacies live in the undercut"); 843 is the lone undermine; 840 is the lone FAIL-LOUD. Cross-PR framing: PR-1's 840 is FAIL-LOUD at the CA-node (RA exists), PR-2's 834/835/837 are FAIL-LOUD at the RA-node (no native comparison scheme). Both FAIL-LOUD layers documented across the two cluster shapes. Representation note (decomposition recorded, not yet serialized) shared. Gate unchanged: proposition only, 0 prod CSV write. Co-Authored-By: Claude-Code <noreply@anthropic.com>

…r (proposition, gated) (#699) * docs(taxonomy): #498 AIF chantier PR-1 — Fallacious comparison cluster (proposition, gated) First cluster of the #498 chantier (ai-01 dispatch yj7u3j primary). Proposition only — GATED, 0 write to prod CSV. Multi-session chantier; this PR advances one coherent cluster. Cluster: 'Fallacious comparison' subfamily (Misleading language family), code=truth from Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv: - pk 839 (Fausse analogie) = MAPPED anchor = the cluster template: Analogy_Inference scheme defeated by DifferencesUndermineSimilarity_Conflict CQ. - 5 unmapped leaves (838, 840-843) modeled as defeasible exceptions to legitimate schemes, in the Walton/AIF paradigm #498 asks for. Proposals (per leaf, scheme + exception/CQ + MappingType): - 841 (Analogie etendue): Analogy_Inference + DifferencesUndermineSimilarity (transitive-chain variant), narrowMatch. - 842 (similarite fallacieuse): Analogy_Inference + DifferencesUndermineSimilarity (mere-resemblance), broadMatch. - 843 (Fausse equivalence): Analogy_Inference + DifferencesUndermineSimilarity (argument-level variant), closeMatch. - 838 (Distinction sans difference): HONEST scheme-divergence — actually a classification fallacy (ArbitraryVerbalClassification_Inference + PropertyNotExistant_Conflict), broadMatch. - 840 (Petition analogique): FAIL LOUD — no AIF native circularity-CQ for analogy exists; documented in AIF_skosOther, no fabricated Conflict token. Coverage: 70 -> 75 mapped (cluster-level). Leaf-level 100% explicitly not the goal (#498 DoD). OWL regen deferred (jsboige ratifies -> CSV edits -> OWL #133). No prod CSV write, no DB write, no regen. Rappels HARD tenus. Co-Authored-By: Claude-Code <noreply@anthropic.com> * fix(taxonomy): #498 PR-1 correct cluster boundary — pk 838 is Faulty comparison, not False analogy Code=truth re-verification against CSV `Soussousfamille` column revealed 3 factual errors in the PR-1 proposition (caught before merge, still in review): 1. pk 838 ("Distinction sans différence") was placed in the False analogy sub-sub — it actually belongs to Faulty comparison. Its honest scheme-divergence modeling (classification fallacy, ArbitraryVerbalClassification_Inference) moves to PR-2. 2. The subfamily root was labeled pk 834 (d3) — the true d2 root is pk 833 "Comparaison fallacieuse" (MAPPED: BiasedClassification_Conflict directRef, skos:broadMatch), which was entirely missing. 3. False analogy leaf count was 5 — it is 4 (840-843). Coverage delta 70→74, not 70→75. Corrections: - TL;DR + §1 table rewritten with exact sub-sub assignments (13 rows, 3 sub-subs, 2 mapped nodes). - §1 adds an explicit "cluster-boundary correction" callout documenting the earlier error. - §3 removes the pk 838 subsection (deferred to PR-2); adds a pointer note. - §4 summary table reduced to 4 leaves (839 anchor + 840-843). - §5/§6 counters reconciled (70→74, 4 leaves, 1/5→5/5 sub-sub). - §5 method notes updated: cluster shape can be anchored by parent d2/d3 when no in-sub-sub anchor. Net effect: PR-1 methodology is strengthened — all 4 False-analogy leaves now honestly reuse Analogy_Inference with no in-cluster scheme-divergence. Gate unchanged: proposition only, 0 prod CSV write. Co-Authored-By: Claude-Code <noreply@anthropic.com> * docs(taxonomy): #498 PR-1 show both 833 Conflict nodes (anchor fidelity for PR-2) pk 833 (d2 root) carries TWO DirectRef Conflict nodes (BiasedClassification_Conflict + ExceptionSimilarityCase_Conflict). The earlier table/TL;DR showed only the first. Corrected to show both, so PR-2 (Faulty comparison, which borrows 833 as anchor and splits its two Conflict nodes across the leaves) can reference 833 faithfully. Co-Authored-By: Claude-Code <noreply@anthropic.com> * docs(taxonomy): #498 PR-1 add §7 enriched-DoD mapping (I-node/RA-node/CA-node) for chantier parity with PR-2 PR-2 #701 carries a §7 mapping each leaf to jsboige's enriched DoD (decompose into I-nodes + RA-node + CA-node, typed undermine/undercut/rebut). PR-1 was missing the parallel section — inconsistent for an ai-01 batch review that will read both PRs together. Added §7 to PR-1: maps the 4 False-analogy leaves to the decomposition. - 841 Analogie étendue → undercut (similarity not transitive; inference defective). - 842 similarité fallacieuse → undercut (resemblance too superficial). - 843 Fausse équivalence → undermine (premise "X ≡ Y" is false). - 840 Pétition analogique → FAIL-LOUD at CA-node level (RA exists, CA missing — circularity has no native AIF Conflict node). Attack-type distribution: 3 of 4 are undercuts (consistent with jsboige's "most fallacies live in the undercut"); 843 is the lone undermine; 840 is the lone FAIL-LOUD. Cross-PR framing: PR-1's 840 is FAIL-LOUD at the CA-node (RA exists), PR-2's 834/835/837 are FAIL-LOUD at the RA-node (no native comparison scheme). Both FAIL-LOUD layers documented across the two cluster shapes. Representation note (decomposition recorded, not yet serialized) shared. Gate unchanged: proposition only, 0 prod CSV write. Co-Authored-By: Claude-Code <noreply@anthropic.com> --------- Co-authored-by: Claude-Code <noreply@anthropic.com>

…(proposition, gated) (#703) Third cluster of the #498 chantier. Completes the Fallacious comparison subfamily (after PR-1 #699 False analogy + PR-2 #701 Faulty comparison). Proposition only — GATED, 0 prod CSV write. Cluster: Association fallacy sub-sub (pk 844-845), no in-sub-sub mapped anchor → borrows the d2 root pk 833 (same borrowed-anchor pattern as PR-2). Both leaves are classification-flavored (like PR-2's 836/838) and honestly fit native classification schemes. 2 leaves, both fully-modeled (cleanest cluster of the chantier, 0 FAIL-LOUD): - 844 Sophisme d'association → VerbalClassification_Inference + BiasedClassification_Conflict (irrelevant shared trait → biased classification), closeMatch. - 845 Amalgame → ArbitraryVerbalClassification_Inference + ExceptionSimilarityCase_Conflict (single common point → exceptional/unrepresentative basis), closeMatch. Anchor-split pattern documented: 844 and 845 cleanly split 833's two Conflict nodes (biased classification vs exceptional similarity) because their semantics differ along that fault line. §7 enriched-DoD: both leaves are undercuts (the invoked association could be real; the inference trait→association is defective). Consistent with jsboige's "most fallacies live in the undercut". Native-vocabulary discipline: all 4 tokens reused from 833/native classification schemes. Validated programmatically — 0 fabrication, 0 FAIL-LOUD this cluster. Cumulative coverage: 76→78 fully-modeled (DoD) / 79→81 DirectRef-loose. Fallacious comparison subfamily now fully addressed (3/3 sub-subs, 13/13 leaves). Co-authored-by: Claude-Code <noreply@anthropic.com>

…oige ratification (#707) Synthesis note (read-only, 0 prod write) consolidating the #498 chantier after 4 PR, to support jsboige ratification of (a) the méthode and (b) the I/RA/CA CSV schema decision. Content: - §1 Coverage code=truth on master 70bd160: 70/1408 (5.0%) per family. Flags a data-quality gap (48 rows with empty Family — out of chantier scope, separate hygiene item). - §2 Chantier progress table: 4 PR, 14 leaves, 70->81 fully-modeled (+11 DoD). Fallacious comparison complete (3/3 sub-subs), Inexact definition opened (1/3). - §2b FAIL-LOUD accounting: 4 leaves across 2 layers (CA-missing PR-1 840, RA-missing PR-2 834/835/837), all honest gaps documented in AIF_skosOther, 0 fabrication. - §3 Méthode recap (cluster selection, 2 patterns exception/direct-conflict, native-vocab discipline #677, fail-loud rules). - §4 DoD enrichi jsboige (I/RA/CA): recorded in each PR §7 but NOT YET SERIALIZED. Decision needed: (a) ratify 2 new CSV columns AIF_attackType + AIF_attackedNode (reco — serialize early avoids late back-fill), (b) defer, (c) alternative schema. - §5 Trajectory: ~87 sub-subs, recommend focus on Misleading language (12.6%) + Faulty logics (10.8%) = best anchor/leaves ratio. Next worker-able targets listed. Derived code=truth from CSV scan + chantier PRs (#699/#701 merged, #703/#705 OPEN). No token fabricated in the synthesis. Co-authored-by: Claude-Code <noreply@anthropic.com>

…t, 7 leaves) (#708) Sixth cluster of the #498 AIF chantier. Models the Inconsistent definition sub-sub (pk 826-832) of the Inexact definition subfamily, borrowing the d2 root pk 799 (exception pattern: VerbalClassification_Inference defeated by BiasedClassification_Conflict + ArbitraryVerbalClassification_Inference). Largest and most varied cluster of the chantier so far — semantically heterogeneous (contradiction / stolen-concept / circularity / lexical rigidity / recursive vagueness / vague meta-fallacy), so the leaves split across three patterns rather than forcing one: - Exception pattern (826 VerbalClassification + Logical_Conflict, 830 VerbalClassification + ArbitraryVerbalClassification) - Direct-conflict pattern (827 Logical_Conflict, 828 InconsistentCommitment, 831 VagueVerbalClassification) - FAIL-LOUD (829 circularity CA-missing like PR-1 840, 832 vague meta-fallacy RA-missing like PR-2 834/835/837) 7 leaves proposed: 5 fully-modeled + 2 FAIL-LOUD (documented in AIF_skosOther, 0 fabrication). Cross-cluster scheme reuse: 831 reuses PR-4's VagueVerbalClassification_Inference. §7 I/RA/CA decomposition recorded but not serialized (pending jsboige schema decision #707 §4). Borrow-root shape = independent of that decision. PROPOSITION ONLY — GATED, 0 prod CSV write, no DB write, no OWL regen. Native vocabulary validated programmatically (all backticked scheme/conflict tokens native; Circularity_Conflict appears only in the "do not fabricate" prohibition, never as a proposal). Relates: dispatch mkzt2v (tertiary), #498, PR-1 #699, PR-2 #701, PR-3 #703, PR-4 #705, coverage-status #707, #133/#130 (OWL), #499, #677 (0 fabrication), #192, #458. Co-authored-by: Claude-Code <noreply@anthropic.com>

…k of 5 clusters (ratification-ready) (#709) Independent read-only audit of po-2023's 5 #498 cluster propositions (PR-1 #699 False analogy, PR-2 #701 Faulty comparison, PR-3 #703 Association fallacy, PR-4 #705 Vague definition, PR-6 #708 Inconsistent definition), triggered by ai-01 dispatch `awhj8g` (PRIMARY) to de-risk jsboige's ratification before any prod CSV write. Method: a deterministic stdlib script (498_audit_extract.py, read-only, scratchpad) extracts the CSV columns for every cited pk (25 pks: 799, 800-803, 826-832, 833-845) and builds the native AIF vocabulary inventory (every *_Inference/*_Conflict token in any AIF column). Verdicts computed by crossing the docs' claims against this extraction — no LLM judgment on the data, only on the semantic legitimacy of FAIL-LOUD cases. Findings — the 5 propositions are code=truth-clean: - 25/25 cited pks present; cluster boundaries match Soussousfamille/Subsubfamily_en for all 5 clusters. - 4/4 mapped anchors (799, 800, 833, 839) match byte-exact on text_fr + DirectRef + ExceptionRef + MappingType. - 10/10 proposed tokens native (inventory of 54 tokens) → 0 fabrication (#677 discipline upheld). - 21/21 cited desc_fr match byte-close (1 cosmetic apostrophe diff on 841). - 6/6 FAIL-LOUD cases legitimate (Comparison_Inference + circularity node genuinely absent from inventory). Two FAIL-LOUD layers verified: CA-missing (840, 829 circularity) + RA-missing (832, 834/835/837). 1 FLAG raised (not a rubber-stamp) — coverage-accounting consistency: - PR-1 counts 840 (CA-missing: RA exists, CA absent) in its "+4 mapped", but PR-6 excludes 829 (same CA-missing shape) from its "+5 fully-modeled". The chantier headline "86 fully-modeled" depends on which PR's convention applies; uniform strict criterion → 85. Recommend jsboige ratify a single criterion. Does not block ratification of the propositions themselves. Secondary (ratification-ready note) integrated as §7 of the audit: verified native-token inventory + the two FAIL-LOUD layers, for jsboige's schema decision (I/RA/CA serialization, per #707 §4). Read-only: 0 write under Cards/, audit doc only. Adversarial (1 FLAG, not rubber-stamp). Dispatch awhj8g (PRIMARY + SECONDARY integrated). Base 34c7702. Co-authored-by: Your <your.email@example.com> Co-authored-by: Claude-Code <noreply@anthropic.com>

#711) Seventh cluster of the #498 AIF chantier. Opens the Ambiguïté subfamily (Abus de language / Misleading language family). Models the Amphibologie sub-sub (pk 847 anchor + 848-854), borrowing the d2 subfamily root pk 846 (direct-conflict pattern: ArbitraryVerbalClassification_Inference + OppositeConsequences_Conflict + SignFromOtherEvents_Conflict). Structurally coherent cluster — all leaves are syntactic/structural/prosodic ambiguity mechanisms (sentence-level), distinct from the Equivoque sub-sub (term-level). This coherence drives an honest scheme-divergence from the borrow-root: amphiboly is syntactic, the borrow-root's classification framing is term-level, so leaves split across native schemes that honestly fit their mechanism rather than forcing the borrow-root's scheme onto syntactic defects: - Exception pattern (849 modal scope, 850 quantifier shift): Deductive_Inference + Logical_Conflict (genuine logical-scope defects, first chantier use of Deductive_Inference for a fallacy) - Direct-conflict pattern (851 accent, 852 illicit contrast): ArbitraryVerbalClassification_Inference (borrowed from 846 — prosody is term-level, classification framing fits) - Direct-conflict + cross-cluster reuse (854 barbarism): VagueVerbalClassification_Inference (PR-4 scheme — barbarism is vagueness not ambiguity; the desc word 'incertain' signals it) - FAIL-LOUD (847 amphiboly anchor, 848 punctuation, 853 solecism): structural/syntactic ambiguity — THIRD recurring AIF ontology gap after circularity (PR-1/PR-6); no native syntactic-ambiguity CQ, documented in AIF_skosOther, 0 fabrication 8 rows proposed: 5 fully-modeled + 3 FAIL-LOUD (most FAIL-LOUD in a single cluster). §7 I/RA/CA decomposition: all typed leaves undercuts (consistent jsboige). Decomposition recorded but not serialized (pending jsboige schema decision #707 §4). Borrow-root shape = independent of that decision. PROPOSITION ONLY — GATED, 0 prod CSV write, no DB write, no OWL regen. Native vocabulary validated programmatically (all backticked scheme/conflict tokens native; Ambiguity_Conflict and Amphiboly_Conflict appear only in the 'do not fabricate' prohibition L105, never as proposals). Relates: dispatch ih617l (primary), #498, PR-1 #699, PR-2 #701, PR-3 #703, PR-4 #705, PR-6 #708, coverage-status #707, verification-audit #709, #133/#130 (OWL), #499, #677 (0 fabrication), #192, #458. Co-authored-by: Claude-Code <noreply@anthropic.com>

…in-sub-sub anchor, 4 leaves) (#713) Eighth cluster of the #498 AIF chantier. Second cluster of the Ambiguïté subfamily (opened by PR-7 Amphibologie). Models the insinuation mechanism of the Ambiguïté narrative sub-sub: in-sub-sub anchor pk 876 (ConflictingGoals_ Conflict, direct-conflict) + 4 leaves (pk 877-880). In-sub-sub anchor shape (cleanest, PR-1/PR-4-like). The sub-sub has 10 leaves spanning 3 mechanisms (insinuation / propaganda / interrogation), so PR-8 scopes to the insinuation mechanism (4 leaves); propaganda + interrogation deferred to PR-9 (different AIF schemes). All 4 leaves are say-without-saying variants — the narrative serves conflicting communicative goals (benign surface + deniable insinuated payload), the anchor 876 mechanism. Two patterns: - Direct-conflict reusing anchor token (877 false implication, 878 innuendo): ConflictingGoals_Conflict (borrowed from 876), narrow/closeMatch - Direct-conflict with semantic-closer native conflict (879 poisoned compliment): OpposedCommitment_Conflict (praise/discredit opposition beats conflicting-goals) - Genuine exception pattern with distinct scheme (880 anonymous sourcing): PositionToKnow_Inference + LackOfCompleteKnowledge_Conflict (epistemic defect, anonymity hides whether source is in position to know) 4 leaves, 4 fully-modeled, 0 FAIL-LOUD (cleanest since PR-3). First chantier use of OpposedCommitment_Conflict and LackOfCompleteKnowledge_Conflict (both native, 2x/1x in existing usage). Palette now 17 distinct native tokens. §6 accounting transparency (re po-2024 audit #709): this PR's +4 is criterion-independent (all leaves have both scheme AND conflict). Cumulative reported under both strict (94) and inclusive (95) criteria — the 1-row discrepancy is the PR-1 baseline (pk 840, RA-exists/CA-missing), pending jsboige criterion decision. §7 I/RA/CA: all 4 undercuts (insinuation defeats the audience's inference, not the surface statement). RA-nodes untyped for 877/878/879 (Gricean implicature — no native Implicature_Inference scheme, 4th ontology gap shape, documented not fabricated). Decomposition recorded, not serialized (in-sub-sub anchor = independent of I/RA/CA schema decision #707 §4). PROPOSITION ONLY — GATED, 0 prod CSV write, no DB write, no OWL regen. Native vocabulary validated programmatically (Implicature_Inference appears only as named gap L281/L283/L303, never as a proposal). Relates: dispatch ih617l (tertiary), #498, PR-1 #699, PR-2 #701, PR-3 #703, PR-4 #705, PR-6 #708, PR-7 #711, coverage-status #707, verification-audit #709, family-gap #712, #133/#130 (OWL), #499, #677 (0 fabrication), #192, #458. Co-authored-by: Claude-Code <noreply@anthropic.com>

…leaves, completes subfamily) (#714) Ninth cluster of the #498 AIF chantier. Completes the Ambiguïté subfamily (3/3 sub-subs). Second half of the Ambiguïté narrative sub-sub: the deferred mechanisms from PR-8 — source-deception (881-883), interrogation (884-885), and polytélie (886). In-sub-sub anchor pk 876 (ConflictingGoals_Conflict). Source-deception and interrogation leaves honestly diverge from the anchor's ConflictingGoals_Conflict to epistemic/action schemes that fit their semantics: - Source-deception (881 grey, 883 black): ExpertOpinion_Inference + ExpertiseInconsistency_Conflict. Omission (concealment, narrow) vs commission (falsification, close) distinguished by MappingType — graded source-deception. - Whisper campaign (882): PositionToKnow_Inference + LackOfCompleteKnowledge_ Conflict (cross-PR-8 reuse of 880 epistemic pair, organized-campaign flavor). - Clandestine interrogation (884): PracticalReasoning_Inference + ConflictingGoals_Conflict (anchor reuse — friendly surface + extraction goal). - Preach the false (885): Bias_Inference direct-conflict (scheme as defect — evidence biased by false-premise collection, degenerate undercut like PR-4 800). - Polytélie (886): ConflictingGoals_Conflict direct reuse (literal multiple goals). 6 leaves, 6 fully-modeled, 0 FAIL-LOUD (2nd consecutive clean cluster). 4 new chantier native tokens: ExpertOpinion_Inference, ExpertiseInconsistency_Conflict, PracticalReasoning_Inference, Bias_Inference (all native 1x/1x/2x/2x). Palette now 21 distinct native tokens across 9 PR. §6 completes Ambiguïté subfamily 3/3 (18/18 leaves). Second fully-complete subfamily after Fallacious comparison. Cumulative 70→100 strict / 101 inclusive (+30). §6 dual criterion re po-2024 audit #709 (this PR's +6 criterion-independent). §7 I/RA/CA: all 6 undercuts (deception defeats the inference audience/target is manipulated to draw). Decomposition recorded, not serialized (in-sub-sub anchor = independent of I/RA/CA schema decision #707 §4). PROPOSITION ONLY — GATED, 0 prod CSV write, no DB write, no OWL regen. Native vocabulary validated programmatically — PASS, 0 fabrication. Relates: dispatch ih617l (tertiary), #498, PR-1 #699, PR-2 #701, PR-3 #703, PR-4 #705, PR-6 #708, PR-7 #711, PR-8 #713, coverage-status #707, verification-audit #709, family-gap #712, #133/#130 (OWL), #499, #677 (0 fabrication), #192, #458. Co-authored-by: Claude-Code <noreply@anthropic.com>

jsboige merged commit 70bd160 into master Jul 5, 2026
3 checks passed

jsboige deleted the docs/498-aif-faulty-comparison-cluster branch July 5, 2026 11:11

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

docs(taxonomy): #498 AIF chantier PR-2 — Faulty comparison cluster (proposition, gated) - #701

docs(taxonomy): #498 AIF chantier PR-2 — Faulty comparison cluster (proposition, gated)#701
jsboige merged 1 commit into
masterfrom
docs/498-aif-faulty-comparison-cluster

jsboige commented Jul 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

jsboige commented Jul 5, 2026

Summary

Cluster shape (differs from PR-1)

5 leaves modeled along a clean semantic fault line

jsboige enriched-DoD compliance (§7)

Native-vocabulary discipline

Honest coverage accounting

Gate (HARD)

Relates

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant