Skip to content

docs(taxonomy): #499 AIF PR-1 (Virtues) — Raisonnement valide / Causalités cluster (proposition, gated) #741

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 7, 2026
Merged

docs(taxonomy): #499 AIF PR-1 (Virtues) — Raisonnement valide / Causalités cluster (proposition, gated) #741

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
108 changes: 108 additions & 0 deletions docs/taxonomy/499-aif-raisonnement-causalite-cluster.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,108 @@
# #499 AIF chantier — PR-1 (Virtues) : Raisonnement valide / Causalités bien identifiées cluster (proposition, gated)

**Issue:** [#499 — Virtues parity (relational + AIF + export)](https://github.com/ArgumentumGames/Argumentum/issues/499) (CLOSED — relational layer done; this advances the **ratified I/RA/CA serialization** #707§4(a))
**Author:** Claude Code @ myia-po-2024 (worker)
**Dispatch:** ai-01 `ompwhx` [primaire] #499 Virtues AIF (schéma #707§4(a) ratifié, CSV distinct Fallacies → 0 collision)
**Date:** 2026-07-07
**Base:** master `defd53d4`
**Status:** **PROPOSITION ONLY — GATED. 0 write to prod CSV.** Mirror of the po-2023 Fallacies chantier format (#498 PR-1 #699). This is the **pilot** for serializing the ratified `AIF_attackType` + `AIF_attackedNode` schema on the Virtues side, with the **inverse-paradigm adaptation** flagged for jsboige validation before any scale-up.

---

## TL;DR

- jsboige ratified schema **#707 §4 = Option (a)** on 2026-07-07 (interactive): two new CSV columns `AIF_attackType` (`undermine`/`undercut`/`rebut`) + `AIF_attackedNode` (`I-node`/`RA-node`/`CA-node`). po-2023 owns the Fallacies side (#498). This PR is the **Virtues mirror pilot**.
- The Virtues relational/AIF layer (12-col `crossLink_*` + `AIF_skos*`) is **already generated and validated CLEAN** (222/222 nodes, 9 PRs merged June 2026, validator #518 12/12 — see [`499-virtues-parity-closure.md`](499-virtues-parity-closure.md)). This PR does **not** redo it; it **extends** it with the I/RA/CA decomposition the ratified schema asks for.
- **One cluster modeled**: *Raisonnement valide* (family 4) / *Causalités bien identifiées* (subfamily 4.1) — anchor pk 80 + leaves 81/82/83. Chosen because it is the **cleanest inverse-paradigm mirror**: Fallacies subfamily *Causalité douteuse* sits at **path 4.1.x** (698 Pétition de principe, 707 Inversion de causalité, 719 Effet cigogne) — the Virtues subfamily *Causalités bien identifiées* sits at **path 4.1**. Same path, opposite tenor.
- **Inverse-paradigm adaptation (FLAGGED, not decided alone):** the ratified columns describe an *attack* (`attackType` on `attackedNode`). A Virtue is the *good holding* of a scheme — it does not attack. Three serialization options exist for the Virtues side (§3); this PR recommends **Option A — "attack resisted"** (the columns record the attack-type the virtue *prevents* on the node it *holds*). Native AIF vocab only (0 fabrication). **jsboige validates the choice** before the 222-row back-fill.

---

## 1. The cluster

**Raisonnement valide** family (Virtues root pk 79, path 4) / **Causalités bien identifiées** subfamily (pk 80, path 4.1). From `Cards/Fallacies/Argumentum Virtues - Taxonomy.csv`:

| pk | depth | path | title_fr | description_fr (excerpt) |
|----|------:|------|----------|--------------------------|
| 80 | 2 | 4.1 | Causalités bien identifiées | Capacité à distinguer clairement les causes des effets… (anchor subfamily) |
| 81 | 3 | 4.1.1 | Indépendance des prémisses | Les prémisses doivent soutenir la conclusion sans la présupposer |
| 82 | 3 | 4.1.2 | Causalité bien orientée | Capacité à reconnaître la direction correcte d'une relation causale |
| 83 | 3 | 4.1.3 | Exclusion des causes alternatives | Analyse méthodique visant à écarter les explications causales concurrentes |

**Opposed Fallacies** (from `Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv`, cross-checked by PK):

| fallacy pk | path | family | title_fr | mechanism (why it is an undercut) |
|-----------|------|--------|----------|-----------------------------------|
| 698 | 4.1.1 | Erreur de raisonnement | Pétition de principe | circularity — the premise presupposes the conclusion → undercuts the RA-node (Rule inference's own support link) |
| 707 | 4.1.2 | Erreur de raisonnement | Inversion de causalité | reverses causal direction → undercuts the RA-node (Cause-to-Effect inference) |
| 719 | 4.1.3 | Erreur de raisonnement | Effet cigogne | correlation≠causation → undercuts the RA-node (Cause-to-Effect inference) |
| 633 | 3.2.1 | Erreur mathématique | Relation infondée | posits a relation without basis → undercuts the RA-node (cross-family, honest scheme-divergence) |

> **Mirror proof.** The Fallacies *Causalité douteuse* subfamily and the Virtues *Causalités bien identifiées* subfamily share the **path prefix 4.1**. The taxonomy's path numbering encodes the inverse pairing at the structure level — independent corroboration of jsboige's inverse paradigm ("a Virtue is the good holding of the same scheme a Fallacy defeats"). The anchor and its 3 leaves carry the same path children (4.1.1/4.1.2/4.1.3) on both sides.

## 2. Reused grounding (no re-derivation)

The 12-col relational layer (merged June 2026) already established, per Virtue PK: the Walton scheme (`AIF_skosDirectRef`), the restored critical question (`AIF_skosMappingType`), and the opposed Fallacy PKs (`crossLink_Opposes`). This PR **reuses** that grounding and only adds the I/RA/CA decomposition:

| Virtue pk | scheme (reused) | restored CQ (reused) | opposed fallacies (reused) |
|----------|-----------------|----------------------|----------------------------|
| 81 | Argument from Rule | Les prémisses soutiennent-elles la conclusion sans la présupposer ? | 698; 1297 |
| 82 | Argument from Cause to Effect | La direction causale invoquée est-elle correctement orientée ? | 707; 719 |
| 83 | Argument from Cause to Effect | Les causes alternatives plausibles ont-elles été écartées ? | 633; 719 |

(source: `docs/taxonomy/499-scaleup-phase2-raisonnement-a-annotations.csv`, rows for pk 81/82/83.)

## 3. The inverse-paradigm adaptation (FLAGGED for jsboige)

The ratified columns are attack-shaped: `AIF_attackType` ∈ {`undermine`, `undercut`, `rebut`} and `AIF_attackedNode` ∈ {`I-node`, `RA-node`, `CA-node`}. A Virtue is not an attack — it is the **good holding** of a scheme. Three ways to serialize the columns on the Virtues side:

| Option | `AIF_attackType` (Virtue side) | `AIF_attackedNode` (Virtue side) | Pros | Cons |
|--------|-------------------------------|----------------------------------|------|------|
| **A — "attack resisted" (recommended)** | the attack-type the virtue **prevents** (inherited from the opposed fallacy: mostly `undercut`) | the node the virtue **holds/defends** (mostly `RA-node`) | Cross-dataset coherent (same native values); encodes the inverse relation; 0 fabrication; useful to the EPITA consumer (knows which attack the virtue is the antidote to) | "attackType" on a virtue is a semantic oxymoron at face value — needs the column doc to say "attack resisted/prevented" |
| B — structurally empty | empty | empty | Honest (a virtue attacks nothing) | Columns useless for 222 nodes; loses the I/RA/CA info the ratified schema exists to capture |
| C — AIF-native support encoding | n/a (AIF has no native "support" attack-type; RA-node *is* the support) | `RA-node` | Conceptually purest (a virtue = an RA-node that holds) | Cannot fill `AIF_attackType` without fabricating a `support` token → violates discipline #677 |

**Recommendation: Option A.** It preserves the ratified native vocabulary (`undercut` + `RA-node` are both native AIF), requires no fabrication, and the cross-dataset coherence is real: the row says "this virtue is the good holding of the RA-node that fallacy X undercuts." The column-level documentation (companion to the prod write) will state that on the Virtues side the pair means *attack resisted*, not *attack committed*.

**This is a judgment call on the inverse-paradigm semantics. It is FLAGGED for jsboige validation before the 222-row back-fill. The pilot below assumes Option A to make the proposition concrete; if jsboige picks B or C, only the last two columns of the table change.**

## 4. Proposition (per leaf, Option A)

I/RA/CA decomposition (Walton/AIF), assuming Option A:

| pk | title_fr | I-node (premises→conclusion, shape) | RA-node (scheme held) | CA-node resisted (attack prevented) | `AIF_attackType` | `AIF_attackedNode` |
|----|----------|-------------------------------------|-----------------------|-------------------------------------|------------------|--------------------|
| 80 | Causalités bien identifiées (anchor) | a causal claim is made and its direction + alternatives are examined | Argument from Cause to Effect | the causal RA-node is held against inversion/single-cause/correlation undercuts | `undercut` | `RA-node` |
| 81 | Indépendance des prémisses | premises are stated and checked for non-circular support of the conclusion | Argument from Rule | the rule RA-node is held against circularity (Pétition de principe 698) | `undercut` | `RA-node` |
| 82 | Causalité bien orientée | a causal relation is asserted and its direction is verified | Argument from Cause to Effect | the causal RA-node is held against direction inversion (Inversion de causalité 707) | `undercut` | `RA-node` |
| 83 | Exclusion des causes alternatives | a cause is identified after ruling out alternatives | Argument from Cause to Effect | the causal RA-node is held against single-cause / spurious-correlation undercuts (Effet cigogne 719, Relation infondée 633) | `undercut` | `RA-node` |

**Distribution:** 4/4 `undercut` on `RA-node` — consistent with jsboige's observation that "most fallacies live in the undercut" (the Virtue mirror correspondingly lives in holding the RA-node against undercuts). No `undermine`, no `rebut` in this cluster (those would surface in other families — e.g. *Honnêteté intellectuelle* resisting evidence-undermining).

## 5. Method (for the chantier's future Virtue PRs)

- **Cluster = one Virtue subfamily** (path X.Y) + its depth-3 leaves. Mirror the po-2023 Fallacies cluster shape (anchor + leaves), one subfamily per PR.
- **Prefer path-matched subfamilies** (like 4.1 here): when the Fallacies side has the same path prefix, the inverse pairing is structurally corroborated — strongest grounding. Otherwise use the `crossLink_Opposes` + `prevented_family_pk` already in the 12-col layer.
- **Reuse the 12-col grounding** (scheme, CQ, opposed PKs) — do not re-derive. The I/RA/CA decomposition only adds: which node the scheme is (RA-node), and which attack-type the opposed fallacy uses (read off the fallacy mechanism: circularity/inversion/correlation → `undercut`; evidence-denial → `undermine`; direct counter-conclusion → `rebut`).
- **Fail-loud** if a Virtue's opposed fallacy has no clean AIF attack-type (e.g. a Virtue opposing a purely presentational fallacy) — document in `AIF_skosOther`, never fabricate an `attackType` value outside {undermine, undercut, rebut}.
- **Honest scheme-divergence** (pk 83 opposes 633 *Relation infondée*, an Erreur mathématique, cross-family): kept; the scheme stays Cause-to-Effect on the Virtue side because that is the scheme the Virtue holds — the cross-family link is recorded in `crossLink_Opposes`, not forced into the scheme.

## 6. FAIL-LOUD / honest gaps (this cluster)

- **0 FAIL-LOUD in this cluster** — every leaf has a native scheme (Cause to Effect / Rule) and a native attack-type (`undercut`). No `AIF_skosOther` needed here.
- **Pending semantic decision** (not a gap, a gate): Option A vs B vs C (§3). The proposition is written under Option A; the back-fill of the 222 rows waits on jsboige's pick.

## 7. Coverage (per #499 + #707§4 DoD)

- This PR: **1 subfamily modeled** (Causalités bien identifiées, 4 nodes: 80 + 81/82/83).
- Virtues total: 222 real nodes across 7 families. At ~3-4 nodes/subfamily and ~1 subfamily/PR, the Virtues I/RA/CA chantier is ~15-20 PRs if done subfamily-by-subfamily — **but** if Option A is ratified, the back-fill is largely **mechanical** (attackType/attackedNode derivable from scheme + opposed-family via the existing 12-col annotations), so a single programmatic back-fill PR is plausible post-validation.
- OWL propagation remains downstream of the prod write (Virtues not yet in `OwlAdapter`) — unchanged from the June closure.

## 8. Gate boundaries (HARD)

- ❌ No prod CSV write, no DB write, no OWL regen, no `Cards/` write. **Proposition doc only.**
- ✅ Reuses the validated 12-col layer (June 2026, #518 CLEAN) — no re-derivation, no fabrication.
- ✅ All AIF tokens native (`undercut`, `RA-node`, `I-node`, `Cause to Effect`, `Rule`) — discipline #677 held.
- ✅ Mirror of po-2023 Fallacies format (proposition gated, 1 cluster, method + flag).

Relates: #499 (Virtues parity, inverse paradigm), #498 (Fallacies chantier, po-203 owner), #707 §4 (ratified I/RA/CA schema, Option a), #677 (0 fabrication), #518 (validator), #133/#130 (OWL), `499-virtues-parity-closure.md`, `499-virtues-prod-write-spec.md`, `498-coverage-status.md`.
Loading