From aa2f6ed173950b89f9646d8ab6dd4a170c70c58f Mon Sep 17 00:00:00 2001 From: Your Date: Wed, 22 Jul 2026 13:36:10 +0200 Subject: [PATCH] =?UTF-8?q?docs(taxonomy):=20#498=20AIF=20cluster=20=C2=AB?= =?UTF-8?q?=20Empoisonner=20le=20puits=20=C2=BB=20=E2=80=94=20third=20unde?= =?UTF-8?q?rmine-dominant=20+=20second=20heterogeneous=20sub-sub=20(taxono?= =?UTF-8?q?my=20finding,=20proposition)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cluster doc Layer C pour la sub-sub « Empoisonner le puits » (Obstruction > Saboter le débat), ancre 1352 sérialisée undermine/I-node FULLY-EMPTY (skos + map + cq vides — plus vide que #845) — 3 leaves source-discrediting (1353 insinuation, 1355 projection, 1356 DARVO) + finding taxo : 4 leaves out-of-mechanism (1354 audience-appeal, 1357 straw-man, 1358 absurdity, 1359 two-wrongs). TROISIÈME cluster UNDERMINE-DOMINANT. Mécanisme = discrédit PRÉVENTIF de la source (infos préjudiciables diffusées avant l'argument). Variante preemptive/ insinuating du pôle undermine (vs #843 self-contradiction directe, #844 attaque personnelle directe). Le pôle undermine a maintenant 3 variantes documentées (le pôle le plus développé). TWIN de Diffamation 1402 (#844). 1402 = defamation RÉACTIVE (attaque réputation en réponse à un argument) ; 1352 = defamation PRÉEMPTIVE (discrédit avant l'argument). Les deux partagent DirectAdHominem_Inference. Ce cluster ferme la paire source-discrediting de la famille Ad-hominem. 3/3 native-fit couche SCHÈME (contrairement à #845 fail-loud scheme) — token DirectAdHominem_Inference natif (ancre 1398 #844) fit poisoning-the-well = ad hominem direct préventif. Héritage légitime. DEUXIÈME sub-sub HÉTÉROGÈNE (après #845). Même discipline « qualify, don't serialize » : 4 leaves misgrouped flaggées pour reclassification (gated jsboige), PAS sérialisées (coverage integrity > count). PREMIER finding CROSS-CLUSTER CORROBORÉ : le straw-man misgrouping apparaît dans #845 (1365 Homme de paille) ET ici (1357 Arguments factices). Deux findings indépendants pointant vers le même grouping straw-man MANQUANT dans la famille Obstruction → signal plus fort pour jsboige. Ancre 1352 FULLY-EMPTY flaggée pour back-fill futur (leaves modélisées plus riches que leur propre ancre — native scheme sur leaves, ancre vide). 3 leaves attack-typed, 0 fabrication (#677) — DirectAdHominem_Inference vérifié natif. Axiome attackType→node-type : 0 violation (3 undermine/I). Coverage 148→151/1408 projeté (live rescan fc5c6349 = 148/1408, propositions docs-only). Sub-sub = 7/7 leaves dispositionnées (3 serialized + 4 out-of-mechanism). PROPOSITION ONLY, 0 écriture prod CSV (post-T&A). Artefacts : - docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md - docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv (4 rows) Idle lane dispatch msg-9akcg4 + ACK msg-3a7ffm (« chaque cluster qui révèle un finding taxo vaut plus qu'un cluster de plus »). Verdict QA = ai-01. Companion à #837/#839 (mixed) + #840 (cigogne) + #841 (accident) + #843/#844 (undermine) + #845 (rebut+finding, straw-man corrobore) + #760 (rebut). Co-Authored-By: Claude-Code --- ...8-aif-empoisonner-le-puits-annotations.csv | 5 + .../498-aif-empoisonner-le-puits-cluster.md | 239 ++++++++++++++++++ 2 files changed, 244 insertions(+) create mode 100644 docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv create mode 100644 docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md diff --git a/docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv b/docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv new file mode 100644 index 00000000..05c904f1 --- /dev/null +++ b/docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv @@ -0,0 +1,5 @@ +fallacy_pk,family,fallacy_name,RA_scheme,attack_type,attacked_component,CA_node,AIF_RA_node,AIF_CA_node,violated_cq,why_not_others,justification +1352,Obstruction,Empoisonnement du puits (anchor),Argument ad hominem (poisoning the well),undermine,I-node / source (prejudicial discrediting),(none — not a counter-conclusion),DirectAdHominem_Inference,(none — not a rebut),Le discrédit préventif de la source (informations préjudiciables diffusées à l'avance) attaque-t-il la crédibilité de la source (CQ de la position-à-savoir / fiabilité de la source) avant même que l'argument ne soit entendu ?,"Ce n'est pas un undercut (la règle d'inférence adverse n'est pas contestée), ni un rebut (pas de contre-conclusion opposée) ; la cible est la source elle-même (I-node), discréditée par des informations préjudiciables pour affaiblir sa position d'avance.","ANCHOR (serialized, skosDirectRef VIDE — fully-empty : atk=undermine/node=I mais skos+map+cq vides). L'empoisonnement du puits undermines la source (I-node) en diffusant des informations préjudiciables pour la discréditer avant tout échange. Schème natif DirectAdHominem_Inference (poisoning the well = ad hominem direct préventif, attesté sur l'ancre 1398 #844). TWIN de Diffamation 1402 (#844). Sub-sub HÉTÉROGÈNE : 3 leaves source-discrediting (1353/1355/1356) + 4 leaves out-of-mechanism (1354 audience-appeal, 1357 straw-man, 1358 absurdity, 1359 two-wrongs) = finding taxo documenté." +1353,Obstruction,Se contenter de poser des questions,Argument ad hominem (insinuation),undermine,I-node / source (insinuated accusation),(none — not a counter-conclusion),DirectAdHominem_Inference,(none — not a rebut),Les questions insidieuses qui suggèrent une faute sans la nommer directement discréditent-elles la source par insinuation (CQ de fiabilité de la source) ?,"Ce n'est pas un undercut (la règle n'est pas contestée), ni un rebut (pas de contre-conclusion) ; la cible est la source (I-node), discréditée par l'insinuation d'une faute.",Spécialisation de l'ancre : les questions insidieuses (leading questions) undermine la source (I-node) par insinuation d'accusation. Schème natif DirectAdHominem_Inference (discrédit de source indirect). Cohérent avec l'ancre undermine/I. Cas distinctif (vecteur = insinuation / suggestion). +1355,Obstruction,Projection psychologique,Argument ad homemin (projection),undermine,I-node / source (projected defects),(none — not a counter-conclusion),DirectAdHominem_Inference,(none — not a rebut),L'attribution à l'adversaire de ses propres défauts discrédite-t-elle la source par projection (CQ de fiabilité de la source) ?,"Ce n'est pas un undercut (la règle n'est pas contestée), ni un rebut (pas de contre-conclusion) ; la cible est la source (I-node), discréditée par l'attribution des défauts du locuteur.",Spécialisation de l'ancre : la projection psychologique undermine la source (I-node) en lui attribuant les propres défauts du locuteur. Schème natif DirectAdHominem_Inference (discrédit de source par projection). Cohérent avec l'ancre undermine/I. Cas distinctif (mécanisme psychologique défensif). +1356,Obstruction,DARVO,Argument ad hominem (reversal),undermine,I-node / source (victim reversal),(none — not a counter-conclusion),DirectAdHominem_Inference,(none — not a rebut),Le déni de l'abus suivi de l'accusation de la victime d'en être responsable discrédite-t-il la source par retournement (CQ de fiabilité de la source) ?,"Ce n'est pas un undercut (la règle n'est pas contestée), ni un rebut (pas de contre-conclusion) ; la cible est la source (I-node — la victime), discréditée par le retournement de culpabilité.","Spécialisation de l'ancre : le DARVO (Deny, Attack, Reverse Victim/Offender) undermines la source (I-node) en inversant les rôles victime/agresseur. Schème natif DirectAdHominem_Inference (discrédit de source par retournement). Cohérent avec l'ancre undermine/I. Cas distinctif (manipulation psychologique — le mécanisme le plus toxique du cluster)." diff --git a/docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md b/docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md new file mode 100644 index 00000000..ddbdec50 --- /dev/null +++ b/docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md @@ -0,0 +1,239 @@ +# 2026-07-22 — #498 AIF chantier : cluster « Empoisonner le puits » (Obstruction > Saboter le débat) — **undermine-dominant + second heterogeneous sub-sub (taxonomy finding)** + +**Scope** : dedicated cluster doc for the **Empoisonner le puits** sub-sub (family **Obstruction**, sub-family +**Saboter le débat**, sub-sub **Empoisonner le puits**) — anchor **1352** (serialized `undermine`/`I-node`, +**fully-empty** — no skos, no mappingType, no CQ) + **3 source-discrediting leaves** (1353, 1355, 1356). **Plus a +taxonomy-quality finding** : the 4 *remaining* leaves (1354, 1357, 1358, 1359) are **out-of-mechanism** (audience +appeal / straw man / absurdity / two-wrongs) — misgrouped fallacies documented for reclassification, **not** +serialized. **Proposition only — GATED, 0 write to prod CSV.** Eighth cluster of the dispatch (ai-01 ACK +`msg-3a7ffm` : *"chaque cluster qui révèle un finding taxo vaut plus qu'un cluster de plus"*). + +**Repo reference** : master `fc5c6349`. Issue : #498. Predecessors : #837/#839 (mixed), #840 (cigogne), #841 +(accident), #843 (first undermine-dominant — Inconsistance), #844 (second undermine-dominant — Attaque +personnelle, *Diffamation 1402 is the twin of this anchor 1352*), #845 (second rebut-dominant — Procès en +incohérence, *first heterogeneous sub-sub*), #760 (first rebut-dominant). + +> **⚠ STATUS = PROPOSITION ONLY.** No prod CSV write. jsboige ratifies → gated apply. + +--- + +## TL;DR + +- **Empoisonner le puits** (Obstruction > Saboter le débat) : anchor **1352** serialized `undermine`/`I-node`, + **fully-empty** (skos + mappingType + CQ all empty — even more empty than #845's attack-columns-only anchor). + **3 source-discrediting leaves proposed** — 1353 Se contenter de poser des questions (insinuation), 1355 Projection + psychologique, 1356 DARVO — all `undermine`/`I-node` with native `DirectAdHominem_Inference` (poisoning the well = + ad hominem direct préventif). +- **Third undermine-dominant cluster.** Mechanism = **preemptive source-discrediting** : the source (I-node) is + discredited by prejudicial information before the argument is even heard. This is the *preemptive/insinuating* + variant of undermine (vs #843 direct self-contradiction, #844 direct personal attack). The undermine-pole now has + 3 documented variants. +- **3/3 native-fit on the scheme layer** (unlike #845, which was fail-loud on the scheme). `DirectAdHominem_Inference` + is natively attested (anchor 1398, #844) and fits poisoning-the-well exactly (ad hominem direct préventif). +- **Second heterogeneous sub-sub (taxonomy finding).** The umbrella "Empoisonner le puits" genuinely covers only + 1352-1353/1355/1356 (source-discrediting). The 4 other rows are **misgrouped** : + - **1354 Convaincre le public** = *audience appeal* (target the audience, not the source) — not source-discrediting. + - **1357 Arguments factices** = *straw man* (fabricate arguments) — not source-discrediting. + - **1358 Utiliser des arguments absurdes** = *absurdity tactic* — not source-discrediting. + - **1359 Répondre par de mauvais arguments** = *two wrongs* (weak counter-argument) — not source-discrediting. + These 4 are flagged for reclassification (gated jsboige), **not serialized** — coverage integrity > coverage count + (same discipline as #845). +- **Twin relationship to #844.** Anchor 1352 (Empoisonner le puits) is the *preemptive* twin of leaf 1402 Diffamation + (in #844 Attaque personnelle, the *reactive* personal attack). Both are ad hominem direct (`DirectAdHominem_Inference`). + This cluster closes the Ad-hominem family's source-discrediting pair. +- **3 attack-typed leaves proposed, 0 fabrication, axiom 0 violation** (3 undermine→I). Projected coverage + **148 → 151 / 1 408** once applied (live rescan master `fc5c6349` = 148/1 408 — propositions are docs-only). + +--- + +## 1. Cluster state (code=truth, master `fc5c6349`) + +From `Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv`, family **Obstruction**, sub-family **Saboter le débat**, +sub-sub **Empoisonner le puits** (8 rows) : + +| pk | text_fr | depth | AIF_attackType | AIF_attackedNode | skos | State | +|----|---------|-------|----------------|------------------|------|-------| +| **1352** | **Empoisonnement du puits** (anchor) | d3 | `undermine` | `I-node` | *(fully empty)* | **✅ serialized (fully-empty)** | +| 1353 | Se contenter de poser des questions | d4 | — | — | — | unmapped — **source-discrediting (this cluster)** | +| 1354 | Convaincre le public plutôt que l'adversaire | d4 | — | — | — | **out-of-mechanism** (audience appeal — finding) | +| 1355 | Projection psychologique | d4 | — | — | — | unmapped — **source-discrediting (this cluster)** | +| 1356 | DARVO | d5 | — | — | — | unmapped — **source-discrediting (this cluster)** | +| 1357 | Arguments factices | d4 | — | — | — | **out-of-mechanism** (straw man — finding) | +| 1358 | Utiliser des arguments absurdes | d5 | — | — | — | **out-of-mechanism** (absurdity — finding) | +| 1359 | Répondre par de mauvais arguments | d5 | — | — | — | **out-of-mechanism** (two wrongs — finding) | + +**Leaf content (code=truth `desc_fr`)** : +- **1352** — « Vous discréditez l'autre partie au moyen d'informations préjudiciables afin d'affaiblir sa position. » +- **1353** — « Vous posez des questions qui insinuent une accusation pour suggérer une faute sans la nommer directement. » +- **1355** — « Vous attribuez à votre adversaire vos propres défauts pour détourner l'attention de vos erreurs. » +- **1356** — « Vous niez l'abus commis, puis vous accusez la victime d'en être responsable ou d'être elle-même l'agresseuse. » + +The sub-sub is **heterogeneous** (same shape as #845) : the source-discrediting umbrella genuinely covers 1352/1353/ +1355/1356 ; the 4 others diverge in mechanism. **This is the finding.** + +--- + +## 2. The anchor model (1352, serialized) — preemptive source-discrediting + +`1352` is serialized `AIF_attackType=undermine`, `AIF_attackedNode=I-node`, but **fully-empty** on +`skosDirectRef`/`skosExceptionRef`/`skosMappingType`/`violatedCq` — even emptier than #845's anchor (which at least +had `skos:broadMatch`). The fallacy attacks the **source** (I-node) by preemptive prejudicial discrediting. This is +the *preemptive/insinuating* variant of undermine — distinct from #843 (direct self-contradiction premise) and #844 +(direct personal attack). The three undermine variants : + +| cluster | variant | how the source (I-node) is discredited | +|---------|---------|----------------------------------------| +| #843 Inconsistance | direct self-contradiction | the source's own claims contradict each other | +| #844 Attaque personnelle | direct personal attack | tone/mocking/insult/gesture against the source | +| **this (Empoisonner le puits)** | **preemptive prejudicial discrediting** | prejudicial info / insinuation / projection / role-reversal against the source | + +**Twin of #844 Diffamation 1402.** Leaf 1402 (Diffamation, in #844) is the *reactive* defamation — attacking the +source's reputation *in response to* an argument. Anchor 1352 (Empoisonner le puits) is the *preemptive* defamation — +discrediting the source *before* the argument is heard. Both share `DirectAdHominem_Inference`. This cluster closes +the Ad-hominem source-discrediting pair. + +**Native scheme fit.** `DirectAdHominem_Inference` (natively attested, anchor 1398 #844) fits poisoning-the-well +exactly : it is an ad hominem direct performed *preemptively*. The 3 source-discrediting leaves inherit this scheme +legitimately (each is a vector of preemptive source-discrediting). **3/3 native-fit on the scheme layer** — unlike +#845 (which was fail-loud on the scheme, native only on the conflict layer). + +--- + +## 3. Proposed AIF structure for the source-discrediting leaves + +All 3 → **undermine / I-node** + native `DirectAdHominem_Inference`. The attacked node is the **source** (I-node), +discredited preemptively. Variety is in the **discrediting vector** (insinuation / projection / role-reversal). + +| pk | text_fr | discrediting vector | attackType | attackedNode | DirectRef | MappingType | +|----|---------|---------------------|-----------|--------------|-----------|-------------| +| 1352 | Empoisonnement du puits (anchor) | prejudicial info (umbrella) | undermine | I-node | *(empty — fully-empty)* | *(empty)* | +| 1353 | Se contenter de poser des questions | insinuation (leading questions) | undermine | I-node | `DirectAdHominem_Inference` (native) | narrowMatch | +| 1355 | Projection psychologique | projection (own defects attributed) | undermine | I-node | `DirectAdHominem_Inference` (native) | narrowMatch | +| 1356 | DARVO | role-reversal (victim→offender) | undermine | I-node | `DirectAdHominem_Inference` (native) | narrowMatch | + +**3 attack-typed leaves proposed, 0 fabrication.** Axiom **0 violation** (3 undermine→I). **3/3 native-fit** on the +scheme layer. **Out-of-mechanism leaves (1354/1357/1358/1359) NOT serialized** — see §4. + +--- + +## 4. Taxonomy-quality finding — the 4 out-of-mechanism leaves + +Same discipline as #845 ("qualify, don't just serialize"). The 4 remaining rows are **misgrouped** — their AIF +mechanism is *not* source-discrediting. Flagged for reclassification (gated jsboige), not serialized. + +| pk | text_fr | its ACTUAL mechanism | why it is NOT source-discrediting | +|----|---------|----------------------|-----------------------------------| +| **1354** | Convaincre le public plutôt que l'adversaire | **audience appeal** (target the audience, not the source) | The target is the *audience*, not the opponent's *credibility*. Distinct mechanism (appeal to audience). | +| **1357** | Arguments factices | **straw man** (fabricate arguments that seem to follow the opponent's logic but are fallacious) | Fabricates arguments to refute — misrepresentation, not source-discrediting. *Note : straw man (1365, flagged in #845) is the sibling misgrouping — 1357 belongs with 1365.* | +| **1358** | Utiliser des arguments absurdes | **absurdity tactic** (present an absurd idea to discomfort the opponent) | A tactical absurdity, not a discrediting of the source. | +| **1359** | Répondre par de mauvais arguments | **two wrongs** (reply to a weak argument with an equally questionable one) | A weak counter-argument (tu-quoque-adjacent on the *argument*, not the *source*). | + +**Recommendation (gated jsboige).** 1354 → audience-appeal family ; 1357 → straw-man family (with #845's 1365) ; +1358/1359 → tactical/counter-argument families. Serialization deferred until reclassification. + +> **Cross-cluster note.** #845 flagged 1365 (Homme de paille = straw man) and this cluster flags 1357 (Arguments +> factices = straw man). These two straw-man leaves are siblings — both belong to a *straw-man sub-sub* that does +> not currently exist in the taxonomy. The two findings corroborate each other and point to a **missing straw-man +> grouping** in the Obstruction family. Flagged for jsboige (taxonomy structure, not just leaf reclassification). + +--- + +## 5. Method notes (additions for the chantier) + +- **Third undermine-dominant cluster — the preemptive variant.** #843 (direct self-contradiction), #844 (direct + personal attack), this (preemptive prejudicial discrediting). The undermine-pole now has 3 documented variants — + the most-developed attack-type pole. The source (I-node) can be discredited by *what it says* (#843), *how it is + attacked* (#844), or *preemptive insinuation* (this). +- **Second heterogeneous sub-sub — the finding pattern repeats.** #845 (Procès en incohérence) was the first + heterogeneous sub-sub ; this is the second. Both reveal the same structural issue : several Obstruction-family + sub-subs mix a core mechanism with misgrouped tactical leaves. The "qualify-don't-serialize" discipline now has 2 + data points — it is becoming a *reliable taxonomy-quality signal*, not a one-off. +- **Cross-cluster straw-man corroboration.** The straw-man misgrouping appears in BOTH #845 (1365) and this (1357). + Two independent findings pointing to the same missing grouping = stronger signal for jsboige. This is the chantier's + first *cross-cluster corroborated finding*. +- **Fully-empty anchor (1352) — a modelling gap to flag.** The anchor has `attackType`/`attackedNode` but *nothing + else* (no skos, no mappingType, no CQ). This is a back-fill candidate for a future anchor-modelling pass — the + leaves here are modelled *richer* than their own anchor (native scheme on leaves, empty anchor). Flagged, not + addressed (anchor re-modelling is gated). + +--- + +## 6. DoD status + +| DoD item (#498) | Status | +|-----------------|--------| +| Effective cluster-level coverage documented | ✅ §1/§3 (3 source-discrediting leaves attack-typed) + §4 (4 out-of-mechanism leaves documented) — sub-sub fully *addressed* (8/8 rows dispositioned) | +| Per unmapped leaf : legitimate decomposition | ✅ §3 (3 source-discrediting : undermine/I + native scheme) + §4 (4 leaves : out-of-mechanism, flagged for reclassification) | +| `AIF_skosMappingType` coherent | ✅ §3 (narrowMatch for the 3 serialized leaves) | +| Fail-loud when no native token fits | ✅ §3 (3/3 native-fit on the scheme — no fail-loud needed) + §4 (out-of-mechanism flag for the 4 misgrouped leaves) | +| OWL regen reflects new structures | ⏳ Deferred — gated apply post-ratification | + +### Coverage accounting (code=truth) + +- **Master coverage (rescan, `fc5c6349`)** : **148/1 408 mapped**. **Axiom 0 violation.** +- **This cluster (proposed)** : **+3 attack-typed** (3 undermine/I) → **151/1 408** projected. The 4 out-of-mechanism + leaves are **not** counted (flagged for reclassification). +- **Obstruction family footprint** : this closes the Ad-hominem source-discrediting pair (preemptive 1352 ↔ reactive + 1402/#844). The Obstruction family now has 4 documented sub-subs (Complication exagérée #837, Attaque personnelle + #844, Procès en incohérence #845, Empoisonner le puits this). + +**Empoisonner le puits sub-sub : fully addressed** (3 serialized + 4 out-of-mechanism documented = 7/7 leaves +dispositioned, + anchor). + +--- + +## 7. Mapping to jsboige's enriched DoD (I-node / RA-node / CA-node decomposition) + +### The 3 serialized leaves → undermine / I-node (preemptive source-discrediting) + +The attacked node is the **source** (I-node), discredited preemptively. The attack is *not* on the inference rule +(undercut) nor a counter-conclusion (rebut). No leaf invokes an RA-node or CA-node as the defeated node. + +#### pk 1353 — Se contenter de poser des questions → **undermine** (insinuation vector) +- **I-node** : **the source** (discredited by insinuated accusation via leading questions). +- **RA-node** : *(none — the rule is not challenged)*. +- **CA-node** : *(none — no counter-conclusion)*. +- **Attack type** : **undermine** (the source's credibility is attacked by insinuation). + +#### pk 1355 — Projection psychologique → **undermine** (projection vector) +- **I-node** : **the source** (discredited by attribution of the speaker's own defects). +- **RA-node** : *(none)*. +- **CA-node** : *(none)*. +- **Attack type** : **undermine** (the source's credibility is attacked by projection). + +#### pk 1356 — DARVO → **undermine** (role-reversal vector) +- **I-node** : **the source / victim** (discredited by reversal of victim/offender roles). +- **RA-node** : *(none)*. +- **CA-node** : *(none)*. +- **Attack type** : **undermine** (the victim's credibility is attacked by role-reversal — the most toxic vector). + +### Cluster attack-type distribution +| pk | discrediting vector | I-node (attacked source) | RA-node | CA-node | Attack type | +|----|---------------------|--------------------------|---------|---------|-------------| +| 1352 | prejudicial info (anchor) | the other party | *(none)* | *(none)* | undermine (serialized) | +| 1353 | insinuation | the source | *(none)* | *(none)* | undermine | +| 1355 | projection | the source | *(none)* | *(none)* | undermine | +| 1356 | role-reversal (DARVO) | the victim | *(none)* | *(none)* | undermine | + +**3/3 undermine / I-node** — uniform attack-type, native scheme (`DirectAdHominem_Inference`), variety in the +discrediting vector (insinuation / projection / role-reversal). + +--- + +## Gate boundaries (HARD — proposition only) + +- ❌ No prod CSV write, no DB write, no OWL regen, no `aif:` assertion emitted. Post-T&A regime. +- ❌ No self-merge. No régén launch. Verdict QA = ai-01. +- ✅ Proposition derived code=truth from the taxonomy CSV (master `fc5c6349`) + `DirectAdHominem_Inference` verified + native (anchor 1398, #844) ; 0 fabricated token (0 fabrication #677). The 4 out-of-mechanism leaves are **not** + serialized (axiom integrity over coverage). +- ✅ Eighth cluster documented (third undermine-dominant — preemptive variant ; **second heterogeneous sub-sub** ; + **first cross-cluster corroborated finding** — straw-man misgrouping in both #845 and this) ; fully-empty anchor + flagged for back-fill ; machine-readable annotation CSV provided for the gated serialization. + +Relates : #498 (chantier), #837 (mixed 1), #839 (mixed 2), #840 (cigogne), #841 (accident), #843 (first +undermine-dominant — Inconsistance), #844 (second undermine-dominant — Attaque personnelle, *Diffamation 1402 = +reactive twin of this anchor 1352*), #845 (second rebut-dominant — Procès en incohérence, *first heterogeneous +sub-sub, straw-man 1365 corroborates this 1357*), #760 (first rebut-dominant), #770 (anchor audit), #763 (OWL AIF +wiring), #677 (0 fabrication), `[[aif-no-inherit-attacktype-from-anchor]]` (discipline — leaves share the anchor's +undermine/source-discrediting mechanism legitimately). Base master `fc5c6349`.