docs(taxonomy): #498 AIF chantier PR-3 — Association fallacy cluster (proposition, gated)

docs/taxonomy/498-aif-association-fallacy-cluster.md

@@ -0,0 +1,209 @@
+# 2026-07-05 — #498 AIF chantier, PR-3 : cluster « Association fallacy » (Fallacious comparison, fin)
+
+**Scope**: third cluster of the #498 chantier — model the **Association fallacy** sub-sub
+(`Soussousfamille=Association fallacy`, pk 844-845) of the Fallacious comparison subfamily. Completes
+PR-1 (False analogy #699) + PR-2 (Faulty comparison #701) → the entire **Fallacious comparison**
+subfamily is now mapped. **Proposition only — GATED, 0 write to prod CSV.** Triggered by ai-01
+dispatch `mkzt2v` (primary).
+
+**Repo reference**: master `70bd1605`. Issue: #498 (reformulated, GO jsboige 2026-06-17 verified).
+Predecessors: PR-1 #699 (False analogy), PR-2 #701 (Faulty comparison).
+
+> **⚠ STATUS = PROPOSITION ONLY.** No prod CSV write, no DB write, no OWL regen. This doc proposes
+> the AIF structure for the cluster's unmapped leaves, derived code=truth from the CSV + the AIF
+> native vocabulary confirmed by existing usage. jsboige ratifies → worker applies the CSV edits in
+> a follow-up PR (gated).
+
+---
+
+## TL;DR
+
+The **Association fallacy** sub-sub has **2 leaves** — pk 844 (d3 anchor "Sophisme d'association") +
+pk 845 (d4 "Amalgame"). As in PR-2, there is **no in-sub-sub mapped anchor** → the cluster borrows
+the d2 root pk 833 ("Comparaison fallacieuse", MAPPED, direct-conflict pattern).
+
+Both leaves are **classification-flavored** (like PR-2's 836/838): they wrongly assign items to a
+category / association on the basis of a shared trait that is either irrelevant (844) or trivially
+singular (845). They honestly reuse native classification schemes — `VerbalClassification_Inference`
+(844) and `ArbitraryVerbalClassification_Inference` (845) — and native Conflict nodes inherited from
+the borrowed anchor 833 (`BiasedClassification_Conflict`, `ExceptionSimilarityCase_Conflict`).
+
+**This is the cleanest cluster of the chantier: 2/2 leaves fully-modeled, 0 FAIL-LOUD.** No missing
+scheme, no missing CQ. It also **completes the Fallacious comparison subfamily** (13 leaves across
+3 sub-subs, all now mapped).
+
+---
+
+## 1. Cluster state (code=truth)
+
+From `Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv`, family **Misleading language**, subfamily
+**Fallacious comparison**, sub-sub **Association fallacy**:
+
+| pk | text_fr | depth | DirectRef | ExceptionRef | MappingType | State |
+|----|---------|-------|-----------|--------------|-------------|-------|
+| **833** | **Comparaison fallacieuse** (borrowed anchor) | d2 | `BiasedClassification_Conflict`, `ExceptionSimilarityCase_Conflict` | — | `skos:broadMatch` | **✅ MAPPED (d2 root, parent of this sub-sub)** |
+| 844 | Sophisme d'association | d3 | — | — | — | unmapped (d3 anchor, **PR-3**) |
+| 845 | Amalgame | d4 | — | — | — | unmapped (**PR-3**) |
+
+The borrowed anchor (833) is the same template as PR-2: a **direct-conflict pattern** (DirectRef to
+two Conflict nodes, ExceptionRef empty). But unlike PR-2's pure-comparison leaves (which had no
+native scheme), both Association-fallacy leaves honestly fit a native classification scheme, so they
+follow PR-1's **exception pattern** (named scheme in ExceptionRef + CQ in DirectRef).
+
+---
+
+## 2. The borrowed anchor (833, recap)
+
+**Pattern**: direct-conflict (no ExceptionRef).
+
+**Conflict nodes** (DirectRef): `BiasedClassification_Conflict` (the invoked association biases the
+classification) + `ExceptionSimilarityCase_Conflict` (the case treated as similar is exceptional, not
+representative).
+
+Both Association-fallacy leaves specialize these: 844 leans on the *biased-classification* side
+(irrelevant trait), 845 leans on the *exceptional-similarity* side (a single common point).
+
+---
+
+## 3. Proposed AIF structure for the leaves
+
+For each leaf: **(a) legitimate scheme, (b) exception/CQ, (c) `AIF_skosMappingType`**. Vocabulary
+restricted to AIF-native tokens confirmed by existing usage.
+
+### pk 844 — Sophisme d'association (Association fallacy / guilt by association, d3 anchor) ✅
+- **desc_fr**: "Vous associez à tort des éléments entre eux en vous fondant sur des caractéristiques
+  communes **sans rapport avec l'argument**."
+- **Legitimate scheme**: `VerbalClassification_Inference` (AIF native) — grouping/classifying items
+  together on the basis of a shared property.
+- **Exception/CQ**: the shared property is **not relevant** to the invoked association — the
+  classification is biased by an irrelevant trait. Native conflict: `BiasedClassification_Conflict`
+  (borrowed from 833).
+- **Proposal**: `ExceptionRef=VerbalClassification_Inference`, `DirectRef=BiasedClassification_Conflict`,
+  `MappingType=skos:closeMatch` (a direct classification-level variant of 833's first Conflict node).
+
+### pk 845 — Amalgame (Conflation) ✅
+- **desc_fr**: "Vous associez à tort une personne ou une chose à une catégorie négative en vous
+  fondant sur **un seul point commun**."
+- **⚠ Semantic note**: the "negative category" framing is emotionally loaded but structurally
+  secondary; the core is *classification on a single, hence unrepresentative, common point*.
+- **Legitimate scheme**: `ArbitraryVerbalClassification_Inference` (AIF native, confirmed 3x in
+  existing usage) — drawing a classification/distinction on an arbitrary (here, singular) basis.
+- **Exception/CQ**: a single common point is an exceptional, non-representative basis for the
+  invoked equivalence. Native conflict: `ExceptionSimilarityCase_Conflict` (borrowed from 833 — the
+  case treated as similar is exceptional, not representative).
+- **Proposal**: `ExceptionRef=ArbitraryVerbalClassification_Inference`,
+  `DirectRef=ExceptionSimilarityCase_Conflict`, `MappingType=skos:closeMatch` (a direct variant of
+  833's second Conflict node).
+
+---
+
+## 4. Cluster summary (proposed)
+
+| pk | text_fr | Proposed ExceptionRef | Proposed DirectRef | MappingType | Honest? |
+|----|---------|----------------------|-------------------|-------------|---------|
+| 844 | Sophisme d'association | `VerbalClassification_Inference` | `BiasedClassification_Conflict` | `skos:closeMatch` | ✅ |
+| 845 | Amalgame | `ArbitraryVerbalClassification_Inference` | `ExceptionSimilarityCase_Conflict` | `skos:closeMatch` | ✅ |
+
+**2 leaves proposed, 2 fully-modeled, 0 FAIL-LOUD.** The cleanest cluster of the chantier. Both
+leaves honestly split the borrowed anchor's two Conflict nodes (844 → biased classification, 845 →
+exceptional similarity), each paired with the native classification scheme that fits its semantics.
+
+---
+
+## 5. Method notes (additions for the chantier, beyond PR-1/PR-2)
+
+- **Anchor-split pattern**: when a borrowed anchor carries **two** Conflict nodes (833), a 2-leaf
+  sub-sub can cleanly assign one Conflict node per leaf if the leaves' semantics differ along the
+  same fault line (844 = irrelevant trait → biased classification; 845 = singular trait →
+  exceptional similarity). This is more honest than forcing both leaves onto the same Conflict.
+- **Emotional-loading is structurally secondary**: 845's "catégorie négative" is a valence feature,
+  not a structural one — the AIF modeling targets the *classification* defect (single arbitrary
+  point), not the negativity. Do not invent a "negative-valence" Conflict node.
+- **Cluster cleanliness signal**: a 0-FAIL-LOUD cluster (this one) vs a high-FAIL-LOUD cluster
+  (PR-2, 3/5) is an honest signal about the AIF ontology's coverage: it covers classification
+  fallacies well, pure-comparison fallacies poorly. Useful for the eventual `Comparison_Inference`
+  ontology-extension decision (PR-2 §7).
+
+---
+
+## 6. DoD status
+
+| DoD item (#498) | Status |
+|-----------------|--------|
+| Effective cluster-level coverage documented (not "100% leaves") | ✅ §4 (76→78 fully-modeled / 79→81 DirectRef-loose; subfamily 13/13 leaves addressed) |
+| Per unmapped leaf: legitimate scheme + exception/CQ (not Latin alone) | ✅ §3 (2 leaves, scheme + CQ each, 0 FAIL-LOUD) |
+| `AIF_skosMappingType` coherent | ✅ §4 (closeMatch for both — direct variants of 833's Conflict nodes) |
+| Fail-loud when no honest scheme fits | N/A this cluster (both leaves have a native scheme) |
+| OWL regen reflects new structures | ⏳ Deferred — jsboige ratifies PR-1 + PR-2 + this PR first, then CSV edits applied, then OWL regen (#133) — not in this PR |
+
+### Coverage accounting (cumulative, chantier-wide)
+| PR | Cluster | Leaves | Fully-modeled (DoD) | DirectRef-loose |
+|----|---------|--------|---------------------|-----------------|
+| baseline | — | — | 70 | 70 |
+| PR-1 #699 | False analogy (839-843) | 4 | 74 (+4) | 74 (+4) |
+| PR-2 #701 | Faulty comparison (834-838) | 5 | 76 (+2: 836, 838) | 79 (+5) |
+| **PR-3 (this)** | **Association fallacy (844-845)** | **2** | **78 (+2)** | **81 (+2)** |
+| **total** | **Fallacious comparison subfamily** | **13** | **78** | **81** |
+
+The **Fallacious comparison subfamily is now fully addressed** (3/3 sub-subs). #498 chantier can
+move to the next Misleading-language subfamily (see PR-3 secondaire / next dispatch).
+
+---
+
+## 7. Mapping to jsboige's enriched DoD (I-node / RA-node / CA-node decomposition)
+
+Per jsboige's enriched DoD (#498 last comment): decompose each fallacy into **I-nodes** (premises +
+conclusion), an **RA-node** (the Walton scheme), and a **CA-node** (the Conflict Application, typed
+undermine/undercut/rebut).
+
+### pk 844 — Sophisme d'association → **undercut**
+- **RA-node**: `VerbalClassification_Inference` — grouping items by a shared property.
+- **I-nodes**: premise P1 "items A and B share property p"; premise P2 "p is classification-relevant
+  to the invoked association"; conclusion C "A and B are associated/equivalent".
+- **CA-node**: `BiasedClassification_Conflict`, applied as an **undercut** on the RA-node — the
+  inference (that sharing p licenses the association) is defective because P2 is false (p is
+  *without rapport* to the argument). The conclusion C may still hold; it is the *inference* that is
+  defective. Classic undercut.
+- **CQ that fails**: "is the shared property relevant to the invoked association?" — no.
+
+### pk 845 — Amalgame → **undercut**
+- **RA-node**: `ArbitraryVerbalClassification_Inference` — classifying an item into a category on an
+  arbitrary (here, singular) basis.
+- **I-nodes**: premise P "X shares a single common point p with category Y"; conclusion C "X ∈ Y".
+- **CA-node**: `ExceptionSimilarityCase_Conflict`, applied as an **undercut** on the RA-node — the
+  inference (that one common point licenses membership in Y) is defective because the case is
+  exceptional/non-representative. Again the inference, not the conclusion, is the target.
+- **CQ that fails**: "is one common point a representative basis for the classification?" — no.
+
+### Cluster attack-type distribution
+| pk | RA-node | CA-node | Attack type |
+|----|---------|---------|-------------|
+| 844 | `VerbalClassification_Inference` | `BiasedClassification_Conflict` | undercut |
+| 845 | `ArbitraryVerbalClassification_Inference` | `ExceptionSimilarityCase_Conflict` | undercut |
+
+**Both leaves are undercuts** — consistent with jsboige's note that "most fallacies live in the
+undercut": the invoked association *could* be real, it is the *inference* (trait → association) that
+is defective. No undermine (no false premise — the shared trait genuinely exists, it is just
+insufficient) and no rebut (the conclusion is not directly denied).
+
+### Representation note
+As in PR-1/PR-2, the I-node/RA-node/CA-node decomposition is **recorded here but not serialized**
+in the CSV — it would require new columns (`AIF_attackType`, `AIF_attackedNode`) and is the terminal
+target of jsboige's comment. Three PRs now demonstrate the decomposition across all three cluster
+shapes (in-sub-sub anchor, borrowed-anchor mixed, borrowed-anchor clean) and all FAIL-LOUD layers
+documented (CA-missing PR-1 840, RA-missing PR-2 834/835/837, none here).
+
+---
+
+## Gate boundaries (HARD — proposition only)
+
+- ❌ No prod CSV write, no DB write, no OWL regen, no `aif:` assertion emitted.
+- ❌ No #674/#596/#666 merge, no régén launch.
+- ✅ Proposition derived code=truth from the taxonomy CSV + AIF vocabulary confirmed by existing
+  usage (no fabricated token — all 4 tokens here are reused from 833 / native classification schemes).
+- ✅ Anchor-split pattern (844 → biased classification, 845 → exceptional similarity) documented, not
+  forced.
+
+Relates: dispatch `mkzt2v` (primary), #498 (chantier), PR-1 #699 (False analogy), PR-2 #701 (Faulty
+comparison), #133/#130 (existing OWL), #499 (inverse: virtue = good tenor of a scheme), #192
+(terminology), #458.