Skip to content

docs(taxonomy): #498 AIF cluster « Empoisonner le puits » — third undermine-dominant + second heterogeneous sub-sub (taxonomy finding, proposition) - #846

Merged
jsboige merged 1 commit into
masterfrom
docs/498-empoisonner-le-puits-cluster
Jul 22, 2026
Merged

docs(taxonomy): #498 AIF cluster « Empoisonner le puits » — third undermine-dominant + second heterogeneous sub-sub (taxonomy finding, proposition)#846
jsboige merged 1 commit into
masterfrom
docs/498-empoisonner-le-puits-cluster

Conversation

@jsboige

@jsboige jsboige commented Jul 22, 2026

Copy link
Copy Markdown
Contributor

What

Idle-lane deliverable (dispatch msg-9akcg4 + ai-01 ACK msg-3a7ffm: "chaque cluster qui révèle un finding taxo vaut plus qu'un cluster de plus"). Eighth Layer C cluster. « Empoisonner le puits » (Obstruction > Saboter le débat). Anchor 1352 serialized undermine/I-node fully-empty (skos + mappingType + CQ all empty). 3 source-discrediting leaves proposed (1353, 1355, 1356) — plus a cross-cluster corroborated taxonomy finding : 4 misgrouped leaves.

The cluster's distinctive character — third undermine-dominant, second heterogeneous sub-sub, first cross-cluster finding

Third undermine-dominant cluster. Mechanism = preemptive source-discrediting (prejudicial info before the argument is 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 — the most-developed attack-type pole.

Twin of #844 Diffamation 1402. Leaf 1402 (#844) = reactive defamation (attack reputation in response to an argument). Anchor 1352 (this) = preemptive defamation (discredit before the argument). Both share DirectAdHominem_Inference. This closes the Ad-hominem source-discrediting pair.

3/3 native-fit on the scheme layer (unlike #845, which was fail-loud on the scheme). DirectAdHominem_Inference (native, anchor 1398 #844) fits poisoning-the-well exactly.

The 3 serialized leaves — variety in the discrediting vector

pk text_fr discrediting vector
1353 Se contenter de poser des questions insinuation (leading questions)
1355 Projection psychologique projection (own defects attributed)
1356 DARVO role-reversal (victim→offender)

All undermine/I-node + native DirectAdHominem_Inference. Axiom 0 violation (3 undermine→I).

Taxonomy-quality finding — 4 misgrouped leaves + cross-cluster corroboration

pk text_fr actual mechanism why NOT source-discrediting
1354 Convaincre le public audience appeal targets the audience, not the source's credibility
1357 Arguments factices straw man fabricates arguments to refute — misrepresentation. Sibling of #845's 1365 (Homme de paille).
1358 Utiliser des arguments absurdes absurdity tactic tactical absurdity, not source-discrediting
1359 Répondre par de mauvais arguments two wrongs weak counter-argument, not source-discrediting

First cross-cluster corroborated finding : the straw-man misgrouping appears in BOTH #845 (1365) and this (1357). Two independent findings pointing to the same missing straw-man grouping in the Obstruction family → stronger signal for jsboige (taxonomy structure, not just leaf reclassification). These 4 are flagged for reclassification, not serialized — coverage integrity > coverage count.

Changes

  1. docs/taxonomy/498-aif-empoisonner-le-puits-cluster.md — cluster doc.
  2. docs/taxonomy/498-aif-empoisonner-le-puits-annotations.csv — 4 rows (anchor + 3 leaves), 12 cols.

Scope / posture

  • PROPOSITION ONLY, 0 prod CSV write (post-T&A). Serialization deferred to gated apply.
  • 0 fabricationDirectAdHominem_Inference verified native ; 4 out-of-mechanism leaves NOT serialized (axiom integrity). Read-only docs lane.
  • Verdict QA = ai-01.
  • Signal : ancres skos=True clean épuisées ; ce cluster est skos-empty AVEC finding taxo (la barre tenue). C'est probablement le dernier cluster à finding avant bascule trad feat(taxonomy): Iterate on non-card taxonomy nodes + adapt GPT-4 enrichment script #141 / dette — j'attends confirmation ai-01 sur la trajectoire.

Refs

🤖 Generated with Claude Code

…ermine-dominant + second heterogeneous sub-sub (taxonomy finding, proposition)

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 fc5c634 = 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 <noreply@anthropic.com>
@jsboige
jsboige merged commit 2becf12 into master Jul 22, 2026
3 checks passed
@jsboige
jsboige deleted the docs/498-empoisonner-le-puits-cluster branch July 22, 2026 14:46
jsboige added a commit that referenced this pull request Jul 22, 2026
…read-only) — MEASURE not verdict (#850)

Acompte compute du dispatch ai-01 (msg-lool0s, bascule po-2024 → audit #847).
Calcule par nœud parent les couches (a)+(b) du method doc
aif-structural-audit-method.md §4, sur 145 fully-modeled + projection 145+8
clusters Layer-C. (c) DEFER (non-instrumenté).

=== Instrument (opérationnel, code=truth) ===
(a) attack-type heterogeneity : Shannon entropy normalisé [0,1] de la
    distribution undercut/undermine/rebut, par family/sub-family/sub-sub.
    a_high = H_norm >= 0.5.
(b) fail-loud rate : fraction de leaves SANS token _Inference natif dans
    skosDirectRef OU skosExceptionRef (prod) / colonnes schème annotation
    (clusters). b_high = >= 0.5.
(c) DEFER : reference-reachability + label-distortion pas instrumentés.

=== HEADLINE ===
145-only : 87uc/53um/5rb | native-scheme 34.5% | fail-loud 65.5% | entropy 0.720
projected (145+46 cluster leaves=191) : 107/72/12 | native 41.9% | fail-loud
58.1% | entropy 0.788

=== Findings (nombres, pas verdict) ===
1. (b) DOMINE : matériel de base matériellement résistant (65.5% fail-loud) →
   hétérogénéité family-level majoritairement axis-B (matériau), pas axis-A
   (regroupement). Quantifie « changes at the margin ».
2. LE CHANTIER CONVERTIT B→A : plier les 8 clusters native-rich fait passer
   Obstruction (b 0.67→0.49) ET Erreur mathématique (b 0.50→0.48) de
   « material-resistance B+A » à « tree-tension A-candidate ». Valide la
   stratégie native-rich du chantier.
3. TREE-TENSION GENUINE (a-high/b-low) concentre sur : Ad hominem (b=0.00→0.21,
   le cluster #844/#845/#846 — specimen tu quoque/whataboutisme du method doc
   §4 devenu nombre), Mauvaise déduction (b=0.20, cluster #843),
   Erreur de raisonnement family (b=0.36→0.27). TOUS c-DEFER (peuvent être
   bridge nodes axis-C, pas défauts).
4. Influence = seule family fully coherent (a-low).

=== Piège corrigé ===
Scan skosDirectRef seul = 92% fail-loud (FAUX). Tokens _Inference vivent dans
skosExceptionRef (43) pas DirectRef (10). Bon dénominateur = union des 2 →
50/145 native-scheme, 65.5% fail-loud. Documenté pour ne pas re-piéger.

=== Gated ===
MESURE pas verdict. Aucun libellé « à réorganiser ». Tags mécaniques only.
Synthèse+verdict = ai-01. 0 prod CSV. Header explicite « INPUT for ai-01
synthesis, GATED jsboige ratification, 0 prod CSV ».

Artefacts :
- docs/taxonomy/847-acompte.csv (145-only, 42 rows)
- docs/taxonomy/847-acompte-projected.csv (145+clusters, 51 rows)
- docs/taxonomy/847-acompte-homogeneity.md (narratif)

Relates #847 #498 #845 #846 #770 #763 #677. Base master 2becf12.

Co-authored-by: Your <your.email@example.com>
Co-authored-by: Claude-Code <noreply@anthropic.com>
jsboige added a commit that referenced this pull request Jul 25, 2026
…ai-01) (#904)

Closes the three-installment measurement chain (#850 layers a+b, #853 layer c
reference-reachability, #857 layer c label-distortion) with the synthesis those
docs explicitly deferred to ai-01.

Verdict: the tree survives the audit. NO restructuring proposed.
- Layer (b) dominates (65.5% fail-loud): most heterogeneity is MATERIAL
  resistance, not authorial grouping error. "Changes at the margin" quantified.
- Ad hominem = cross-cultural BRIDGE node (sole `borrowed` of 39). c-DEFER
  lifted; annotate the seam, do NOT flatten. Ratifies #845's deferred decision.
- Axis-C lexical pressure on the grouping layer is LOW (1/39) and the EN
  grouping tree is a 1:1 FR calque -> tradition divergence lives at the LEAF
  reference level, not in the groupings.
- Applies a minimum-n confidence gate the acompte's n>=3 threshold lacks:
  6 axis-A candidates reduce to 2 for jsboige (Biais naturels, Pensee biaisee).
  Argument bacle is EVIDENCE-BLOCKED (46% wiki / 22% longtail) -> repair refs
  before arbitrating. Erreur de raisonnement watch-only; 2 more thin at n=4-5.
- The chantier works, measurably: folding native-rich clusters moves fail-loud
  -7.4 pp and flips Obstruction + Erreur mathematique from B+A to A-candidate.
  Next clusters: Abus de langage, Tricherie, Insuffisance.
- Reference-repair backlog (S5) is actionable WITHOUT jsboige.

Records the honest caveat that `unflagged` is the absence of a flag, never a
positive scope match: the stretched/narrowed half of the axis-C constraint is
unmeasured and mechanically unmeasurable (both proxies failed, documented #857).

VERDICT, not application. 0 prod-CSV write. All structural change stays gated
on jsboige ratification.

Refs #847 #498 #845 #846 #850 #853 #857

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant