Skip to content

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

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jsboige merged 1 commit into from
Jul 5, 2026
Merged

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

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
277 changes: 277 additions & 0 deletions docs/taxonomy/498-aif-faulty-comparison-cluster.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,277 @@
# 2026-07-05 — #498 AIF chantier, PR-2 : cluster « Faulty comparison » (Fallacious comparison, suite)

**Scope**: second cluster of the #498 chantier — model the **Faulty comparison** sub-sub
(`Soussousfamille=Faulty comparison`, pk 834-838) of the Fallacious comparison subfamily. Continues
PR-1 (False analogy sub-sub, #699). **Proposition only — GATED, 0 write to prod CSV.** Triggered by
ai-01 dispatch `yj7u3j` (tertiary/idle: « avance #498 grappe suivante »).

**Repo reference**: master `dc02e847`. Issue: #498 (reformulated, GO jsboige 2026-06-17 verified).
Predecessor: PR-1 (#699, False analogy sub-sub).

> **⚠ 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 (26 Conflict nodes / 36 Inference schemes across
> the 70 mapped rows). jsboige ratifies → worker applies the CSV edits in a follow-up PR (gated).

---

## TL;DR

The **Faulty comparison** sub-sub has **no in-sub-sub mapped anchor** — unlike PR-1's False analogy
(which had pk 839). It is therefore **anchored by the borrowed d2 root pk 833** "Comparaison
fallacieuse" (MAPPED, `skos:broadMatch`). 833 establishes a **different AIF pattern** from PR-1's
anchor: a **direct-conflict pattern** (DirectRef to `BiasedClassification_Conflict` +
`ExceptionSimilarityCase_Conflict`, ExceptionRef empty) rather than PR-1's exception pattern
(ExceptionRef to `Analogy_Inference`).

This PR proposes the AIF modeling for the **5 leaves** (834 d3 anchor + 835/836/837/838 d4). The
cluster splits cleanly along a **semantic fault line**:

- **2 classification leaves (honest, fully modeled)** — 836 + 838 are genuinely *classification*
fallacies (overlapping categories / distinction without difference) and reuse the native
`VerbalClassification_Inference` / `ArbitraryVerbalClassification_Inference` schemes.
- **3 comparison leaves (FAIL-LOUD on the scheme)** — 834/835/837 are *pure comparison* failures
(abusive / incomplete / incoherent comparison) for which **no native AIF Inference scheme exists**
(verified: the 36-scheme vocabulary has no `Comparison_Inference`). They take a DirectRef
(`ExceptionSimilarityCase_Conflict`, borrowed from 833) but their ExceptionRef is documented as
absent rather than fabricated.

**Net**: a higher FAIL-LOUD rate (3/5) than PR-1 (1/4) — an honest signal that the AIF ontology
covers analogy (PR-1) better than pure comparison (this PR).

---

## 1. Cluster state (code=truth)

From `Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv`, family **Misleading language**, subfamily
**Fallacious comparison**, sub-sub **Faulty comparison**:

| 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)** |
| 834 | Comparaison abusive | d3 | — | — | — | unmapped (d3 anchor, **PR-2**) |
| 835 | Comparaison incomplète | d4 | — | — | — | unmapped (**PR-2**) |
| 836 | Classification non exclusive | d4 | — | — | — | unmapped (**PR-2**) |
| 837 | Comparaison incohérente | d4 | — | — | — | unmapped (**PR-2**) |
| 838 | Distinction sans différence | d4 | — | — | — | unmapped (**PR-2**) |

The borrowed anchor (833) is the **template**, but its pattern differs from PR-1's anchor (839):

- **839 (PR-1)** = exception pattern: a legitimate scheme (`Analogy_Inference`) is named in
ExceptionRef, defeated by a violated CQ (`DifferencesUndermineSimilarity_Conflict`) in DirectRef.
- **833 (PR-2 anchor)** = direct-conflict pattern: no ExceptionRef (no legitimate scheme named);
DirectRef carries two Conflict nodes directly. The fallacy *is* the conflict.

This PR honors both patterns: the comparison leaves (834/835/837) follow 833's direct-conflict
shape (DirectRef only), while the classification leaves (836/838) follow 839's exception shape
(named scheme + CQ) because a native scheme honestly fits.

---

## 2. The borrowed anchor (833, the template)

**Pattern**: direct-conflict (no ExceptionRef).

**Conflict nodes** (DirectRef): `BiasedClassification_Conflict` (the comparison biases the
classification) + `ExceptionSimilarityCase_Conflict` (the invoked similarity is an exceptional case,
not representative).

**Mapping type**: `skos:broadMatch` — "Comparaison fallacieuse" is the broad informal-logic name;
the two Conflict nodes are the specific AIF mechanisms.

**Why two Conflict nodes**: a fallacious comparison can fail because (a) the classification it
rests on is biased (`BiasedClassification_Conflict`) or (b) the case treated as similar is actually
exceptional (`ExceptionSimilarityCase_Conflict`). 833 carries both; the leaves specialize.

---

## 3. Proposed AIF structure for the leaves

For each leaf: **(a) legitimate scheme (or FAIL-LOUD), (b) exception/CQ or direct conflict,
(c) `AIF_skosMappingType`**. Vocabulary restricted to AIF-native tokens confirmed by existing usage.

### pk 834 — Comparaison abusive (Abusive comparison, d3 anchor)
- **desc_fr**: "Vous utilisez une comparaison excessive ou inappropriée pour appuyer votre argument."
- **Legitimate scheme**: ⚠ **FAIL LOUD** — no native AIF Inference scheme models "comparison" as a
legitimate presumptive scheme. The 36-scheme vocabulary has no `Comparison_Inference`; the closest
(`VerbalClassification_Inference`, `Example_Inference`, `Sign_Inference`) do not honestly capture
*comparison* as the operative inference.
- **Direct conflict**: `ExceptionSimilarityCase_Conflict` (borrowed from 833 — the comparison treats
an excessive/inappropriate case as representative).
- **Proposal**: `DirectRef=ExceptionSimilarityCase_Conflict`, `ExceptionRef=` *(absent — FAIL LOUD)*,
`AIF_skosOther="No native AIF Inference scheme for non-faulty comparison; DirectRef-only
(direct-conflict pattern, mirroring anchor 833)"`, `MappingType=skos:broadMatch`. Do **not**
fabricate a `Comparison_Inference` token.

### pk 835 — Comparaison incomplète (Incomplete comparison)
- **desc_fr**: "Vous faites une comparaison en omettant des informations importantes, ce qui rend
votre argument peu clair ou incorrect."
- **Legitimate scheme**: ⚠ **FAIL LOUD** — same gap as 834 (no native comparison scheme).
- **Direct conflict**: `ExceptionSimilarityCase_Conflict` (the comparison rests on an incomplete,
hence unrepresentative, case).
- **Proposal**: `DirectRef=ExceptionSimilarityCase_Conflict`, `ExceptionRef=` *(absent — FAIL LOUD)*,
`AIF_skosOther="No native AIF scheme for non-faulty comparison; DirectRef-only"`,
`MappingType=skos:broadMatch`.

### pk 836 — Classification non exclusive (Non-exclusive classification) ✅ honest
- **desc_fr**: "Vous utilisez des catégories qui se chevauchent, faussant ainsi votre comparaison."
- **⚠ Semantic shift**: despite sitting in the comparison sub-sub, this leaf is explicitly a
*classification* fallacy (overlapping categories). It honestly targets a classification scheme.
- **Legitimate scheme**: `VerbalClassification_Inference` (AIF native, confirmed in existing usage).
- **Exception/CQ**: `BiasedClassification_Conflict` (borrowed from 833 — the non-exclusive categories
bias the classification).
- **Proposal**: `ExceptionRef=VerbalClassification_Inference`, `DirectRef=BiasedClassification_Conflict`,
`MappingType=skos:closeMatch` (a direct classification-level variant of 833's first Conflict node).

### pk 837 — Comparaison incohérente (Incoherent comparison)
- **desc_fr**: "Vous comparez des choses en ne retenant que certains aspects, ce qui fausse la
comparaison globale."
- **Legitimate scheme**: ⚠ **FAIL LOUD** — same gap (no native comparison scheme).
- **Direct conflict**: `ExceptionSimilarityCase_Conflict` (cherry-picked aspects → unrepresentative
case).
- **Proposal**: `DirectRef=ExceptionSimilarityCase_Conflict`, `ExceptionRef=` *(absent — FAIL LOUD)*,
`AIF_skosOther="No native AIF scheme for non-faulty comparison; DirectRef-only"`,
`MappingType=skos:broadMatch`.

### pk 838 — Distinction sans différence (Distinction without difference) ✅ honest scheme-divergence
- **desc_fr**: "Vous distinguez deux éléments qui sont en réalité semblables ou qui ne présentent
aucune différence significative."
- **⚠ Mirror fallacy** (moved here from PR-1's earlier draft — see PR-1 §1 boundary correction):
claiming a distinction where none exists. It is a *classification* failure, not an analogy failure.
- **Legitimate scheme**: `ArbitraryVerbalClassification_Inference` (AIF native, confirmed 3x in
existing usage).
- **Exception/CQ**: `PropertyNotExistant_Conflict` (the distinguishing property does not obtain —
confirmed 3x in existing usage).
- **Proposal**: `ExceptionRef=ArbitraryVerbalClassification_Inference`,
`DirectRef=PropertyNotExistant_Conflict`, `MappingType=skos:broadMatch`. **This is the honest
scheme** — 838 is a classification fallacy, and it sits in the comparison subfamily only because
the distinction it attacks is comparison-adjacent.

---

## 4. Cluster summary (proposed)

| pk | text_fr | Proposed ExceptionRef | Proposed DirectRef | MappingType | Honest? |
|----|---------|----------------------|-------------------|-------------|---------|
| 834 | Comparaison abusive | *(absent — FAIL LOUD)* | `ExceptionSimilarityCase_Conflict` | `skos:broadMatch` | ⚠ FAIL LOUD (no native scheme) |
| 835 | Comparaison incomplète | *(absent — FAIL LOUD)* | `ExceptionSimilarityCase_Conflict` | `skos:broadMatch` | ⚠ FAIL LOUD (no native scheme) |
| 836 | Classification non exclusive | `VerbalClassification_Inference` | `BiasedClassification_Conflict` | `skos:closeMatch` | ✅ |
| 837 | Comparaison incohérente | *(absent — FAIL LOUD)* | `ExceptionSimilarityCase_Conflict` | `skos:broadMatch` | ⚠ FAIL LOUD (no native scheme) |
| 838 | Distinction sans différence | `ArbitraryVerbalClassification_Inference` | `PropertyNotExistant_Conflict` | `skos:broadMatch` | ✅ (scheme-divergence, honest) |

**5 leaves proposed: 2 fully-modeled (836, 838) + 3 FAIL-LOUD (834/835/837).** The 3 FAIL-LOUD take
a DirectRef (direct-conflict pattern, mirroring anchor 833) but document the absence of a native
comparison Inference scheme in `AIF_skosOther` rather than fabricating one.

**Honest accounting**:
- If "mapped" = has a DirectRef (the CSV's looser criterion, which counts 833 itself as mapped):
this PR adds 5 → coverage 74→79.
- If "mapped" = satisfies #498 DoD (legitimate scheme + CQ, not DirectRef-only): this PR adds 2
fully-modeled + 3 partial (documented gap) → coverage 74→76 fully-modeled, with 3 honestly
deferred to the `AIF_skosOther` track.

Both counts are reported; the DoD-honest number (76) is the one that matters for #498.

---

## 5. Method notes (additions for the chantier, beyond PR-1)

- **Borrowed-anchor cluster shape**: when a sub-sub has no in-sub-sub mapped anchor (Faulty
comparison), it borrows the nearest mapped ancestor (d2 root 833). The borrowed anchor's
**pattern** (direct-conflict vs exception) dictates the leaves' shape. This is the second cluster
shape of the chantier (PR-1 = in-sub-sub anchor + exception pattern; PR-2 = borrowed anchor +
mixed pattern).
- **Direct-conflict vs exception pattern**: an anchor with ExceptionRef empty (833) models the
fallacy *as* the conflict; an anchor with a named scheme in ExceptionRef (839) models it *as a
defeater of* the scheme. Leaves should follow their anchor's pattern where the semantics allow,
and shift to the exception pattern (named scheme + CQ) only when a native scheme honestly fits
(836, 838).
- **FAIL-LOUD on scheme, not on conflict**: where a native Inference scheme is absent (pure
comparison), still assign the honest Conflict node in DirectRef (don't fail loud on both) — the
conflict is real, only the scheme is missing. Document the missing scheme in `AIF_skosOther`.
- **Semantic-shift leaf**: a leaf's sub-sub placement is not always its true scheme (836 is a
classification fallacy in a comparison sub-sub; 838 too). Reuse the right scheme when the desc_fr
clearly indicates it; don't force the sub-sub's nominal theme.
- **Native-vocabulary discipline**: restrict proposals to AIF tokens confirmed by existing usage
(the 26 Conflict nodes / 36 Inference schemes observed in the 70 mapped rows). Never fabricate a
`*_Conflict` or `*_Inference` token — FAIL LOUD instead.

---

## 6. DoD status

| DoD item (#498) | Status |
|-----------------|--------|
| Effective cluster-level coverage documented (not "100% leaves") | ✅ §4 (74→76 fully-modeled / 79 DirectRef-loose, cluster-level) |
| Per unmapped leaf: legitimate scheme + exception/CQ (not Latin alone) | ✅ §3 — 2 leaves fully (836, 838); 3 leaves FAIL-LOUD on scheme (834/835/837), DirectRef-only documented |
| `AIF_skosMappingType` coherent (broadMatch when looser) | ✅ §4 (broadMatch for DirectRef-only + scheme-divergence; closeMatch for the direct variant 836) |
| Fail-loud when no honest scheme fits | ✅ §3 pk 834/835/837 (no native comparison scheme — documented, not fabricated) |
| OWL regen reflects new structures | ⏳ Deferred — jsboige ratifies PR-1 + this PR first, then CSV edits applied, then OWL regen (#133) — not in this PR |

---

## 7. Mapping to jsboige's enriched DoD (I-node / RA-node / CA-node decomposition)

The last jsboige comment on #498 enriches the DoD: modeling a fallacy "AIF-style" is not just
`scheme + exception-name` — it **decomposes the argument** into **I-nodes** (premises + conclusion),
an **RA-node** (the inference principle = the Walton scheme, a first-class attackable object), and a
**CA-node** (the Conflict Application, typed by attack: **undermine** a premise / **undercut** the
RA-node / **rebut** the conclusion). Most fallacies live in the **undercut**. This section maps the
2 fully-modeled leaves to that decomposition, and documents where the 3 FAIL-LOUD leaves fall short.

### pk 836 — Classification non exclusive → **undercut**
- **RA-node** (inference principle): `VerbalClassification_Inference` — classifying an item into a
category on the basis of its properties.
- **I-nodes**: premise P1 "categories X and Y are mutually exclusive"; premise P2 "item A has the
defining property of X"; conclusion C "A ∈ X (not Y)".
- **CA-node**: `BiasedClassification_Conflict`, applied as an **undercut** on the RA-node — the
inference (that A can be *cleanly* classified) is defective because P1 is false (the categories
overlap). The conclusion C may still hold; it is the *inference* that is defective, not the
conclusion. Classic undercut.
- **CQ that fails**: "are the categories mutually exclusive?" — no (they overlap).

### pk 838 — Distinction sans différence → **undermine**
- **RA-node**: `ArbitraryVerbalClassification_Inference` — drawing a distinction between two classes
on the basis of a distinguishing property.
- **I-nodes**: premise P1 "A has property p that B lacks"; premise P2 "p is classification-relevant";
conclusion C "A and B are in different classes".
- **CA-node**: `PropertyNotExistant_Conflict`, applied as an **undermine** on P1 — the premise "A
has p that B lacks" is false (p does not obtain; A and B are indistinguishable on p). Unlike 836,
the attack lands on a *premise* (I-node), not the inference itself.
- **CQ that fails**: "does the distinguishing property p actually obtain?" — no.

### pk 834/835/837 — FAIL-LOUD at the **RA-node level**
- These leaves cannot be decomposed because the **RA-node is missing**: there is no native AIF
Inference scheme for "comparison" (verified — the 36-scheme vocabulary has no
`Comparison_Inference`). Without an RA-node there is nothing to undercut and no inference to
attack; only a dangling CA-node (`ExceptionSimilarityCase_Conflict`) survives.
- This is a deeper gap than a missing CQ: it is a missing *scheme*. The enriched DoD makes the gap
sharper — FAIL-LOUD here is not "no CQ" (PR-1's 840) but "no RA", which is structurally prior.
- **Terminal remedy (out of scope, jsboige decision)**: either (a) extend the AIF vocabulary with a
`Comparison_Inference` scheme (ontology change, requires ratification), or (b) re-express these
leaves as `Example_Inference` / `Sign_Inference` defeaters (semantic stretch, lower fidelity).
Both are gated; this PR documents the gap, does not pick.

### Representation note
This PR (and PR-1) record the modeling in the **existing CSV columns** (`AIF_skosDirectRef` =
CA-node conflict, `AIF_skosExceptionRef` = RA-node scheme), consistent with the 70 already-mapped
rows. The enriched I-node/RA-node/CA-node decomposition above is **not yet serialized** — it would
require new CSV columns (e.g. `AIF_attackType`, `AIF_attackedNode`) and is the terminal target
flagged in jsboige's comment. Recording the decomposition here (in the proposition doc) lets jsboige
ratify the CA-node typing *before* any CSV schema evolution. PR-1's leaves can be back-filled with
the same decomposition in a follow-up once the column schema is agreed.

---

## Gate boundaries (HARD — proposition only)

- ❌ No prod CSV write, no DB write, no OWL regen, no `aif:` assertion emitted.
- ❌ No #674/#596 merge, no régén launch.
- ✅ Proposition derived code=truth from the taxonomy CSV + AIF vocabulary confirmed by existing
usage (no fabricated token — `Comparison_Inference` absent → FAIL LOUD, not invented).
- ✅ Honest scheme-divergence on 836/838 (classification leaves in a comparison sub-sub); FAIL-LOUD
on the scheme for the 3 pure-comparison leaves, with DirectRef preserved.

Relates: dispatch `yj7u3j` (tertiary/idle), #498 (chantier), PR-1 #699 (False analogy sub-sub,
predecessor), #133/#130 (existing OWL), #499 (inverse: virtue = good tenor of a scheme), #192
(terminology), #458.
Loading