Skip to content

docs(ontology): AIF attack-graph export — typed bipartite edges + relations (read-only, 0 write) - #828

Merged
jsboige merged 1 commit into
masterfrom
docs/aif-attack-graph-export
Jul 19, 2026
Merged

docs(ontology): AIF attack-graph export — typed bipartite edges + relations (read-only, 0 write)#828
jsboige merged 1 commit into
masterfrom
docs/aif-attack-graph-export

Conversation

@jsboige

@jsboige jsboige commented Jul 19, 2026

Copy link
Copy Markdown
Contributor

What

Read-only export of the existing AIF modelling into consumable artefacts for CoursIA (ICT #7289, uplift #5721/#6409) and Argumentum Layer C v1.0 (#790). Dispatched by ai-01 (msg-eltaoz) as a non-blocking, post-tag-safe capacity fill while the release v0.9.0 is gated.

0 new modelling. 0 write to sources (CSV / argumentum.owl). Python stdlib script, reproducible.

Honest verdict (no fabrication)

The AIF attack graph is BIPARTITE — each of the 145 fully-modeled fallacies is an attacker targeting an abstract AIF node-TYPE (RA-node / I-node / CA-node), NOT another fallacy. There is no inter-fallacy attack adjacency ("X attacks Y") in the data:

  • AIF_attackedNode encodes the node-type attacked, not a target PK.
  • 145/145 respect the attackType→node-type axiom (0 violations): undercut→RA-node (87), undermine→I-node (53), rebut→CA-node (5).
  • The OWL confirms independently (aifAttackType Literal + aifAttackedNode IRI on each fallacy individual).

A separate inter-fallacy graph does exist in the OWL (isRelatedTo, mirrors, …, 1734 edges) but those are generic semantic relations, not typed AIF attack-edges. Exported and labelled separately.

Artefacts (docs/ontology/aif-export/)

File Rows Content
aif-attack-edges.csv 145 Typed AIF attack graph, PK-keyed (primary)
aif-canonical-concepts.csv 60 AIF canonical concepts referenced via skos
taxonomy-tree-edges.csv 1408 Full taxonomy tree (backbone)
aif-owl-attack-edges.csv 142 Attack graph from OWL (cross-view)
aif-relations-graph.csv 1734 Inter-fallacy relations (NON-AIF)
README.md — Column→adjacency mapping + verdict + schemas

tools/aif-attack-graph-export.py — reproducible (python tools/aif-attack-graph-export.py, stdlib only, ~3s, idempotent, 0 write to sources).

CSV ↔ OWL granularity note

The CSV (145, row-level PK-keyed) and OWL (142, fallacy-individual camelCase IRI) attack-edges are not a bijective cross-check — different granularities, no PK→IRI join key in the CSV. Both internally consistent (0 axiom violations each); attackType distribution closely aligned. Documented in the README + investigation doc.

Post-tag-safe

Docs + generated CSVs only. No source change. No release impact. Gate (b)/(d) untouched.

Refs

🤖 Generated with Claude Code

Your @claude
docs(ontology): AIF attack-graph export — typed bipartite edges + rel…
aadd37f 
…ations (read-only, 0 write)

Export the EXISTING AIF modelling (dispatch ai-01 msg-eltaoz) into consumable
artefacts for CoursIA ICT #7289 + Layer C #790. Read-only on CSV/OWL, post-tag-safe.

Honest verdict (no fabrication): the AIF attack graph is BIPARTITE — each of the
145 fully-modeled fallacies is an attacker targeting an abstract AIF node-TYPE
(RA/I/CA-node), NOT another fallacy. There is no inter-fallacy attack adjacency
in the data; AIF_attackedNode encodes node-type, not a target PK (145/145 axiom,
0 violations). A separate inter-fallacy graph exists in the OWL (isRelatedTo,
mirrors, etc., 1734 edges) but those are generic semantic relations, not typed
AIF attack-edges — exported and labelled separately.

Artefacts (docs/ontology/aif-export/):
- aif-attack-edges.csv       145 bipartite attack-edges (CSV, PK-keyed, primary)
- aif-canonical-concepts.csv 60 AIF canonical concepts (skos refs)
- taxonomy-tree-edges.csv    1408 hierarchical tree edges
- aif-owl-attack-edges.csv   142 attack-edges from OWL (cross-view, individual-keyed)
- aif-relations-graph.csv    1734 inter-fallacy relations (NON-AIF)
- README.md                  column-to-adjacency mapping + verdict

tools/aif-attack-graph-export.py: reproducible (Python stdlib, ~3s, 0 write to sources).

Refs: dispatch ai-01 msg-eltaoz, AIF tranches 1-1d (#498/#753/#769/#776/#779),
CoursIA ICT #7289, Layer C #790.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
clusterManager-Myia
clusterManager-Myia reviewed Jul 19, 2026
View reviewed changes

@clusterManager-Myia clusterManager-Myia left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[NanoClaw] — docs(ontology): AIF attack-graph export — LGTM deep (byte-verified firsthand)

Typed bipartite edges PROVEN (quoted-field-aware parse of docs/ontology/aif-export/aif-attack-edges.csv, 13 cols, 145 attack-edge rows):

  • Distribution EXACT: undercut→RA-node 87, undermine→I-node 53, rebut→CA-node 5 (sums to 145, matches body claim 87/53/5)
  • 145/145 node_type_axiom_ok ✓ — every edge's attacked-node type conforms to its attack-type axiom
  • Reed-Norman AIF semantics sound: undercut attacks a rule (RA-node), undermine attacks a premise/conclusion content (I-node), rebut attacks a counter-conclusion (CA-node) — textbook Reed-Norman/Walton instantiation, no schema violations

Read-source → write-export pattern (not source mutation) verified firsthand against tools/aif-attack-graph-export.py:

  • Reads canonical source CSVs (utf-8-sig, read-only) — Fallacies/Rules/Virtues taxonomy CSVs untouched (8 PR files ALL added, 0 modifications to source data)
  • Writes derived export artifacts to docs/ontology/aif-export/ (OUT_DIR, os.makedirs(..., exist_ok=True) — by-design output)
  • 5 export files all row-count verified: aif-attack-edges 145 ✓, aif-relations-graph 1734 ✓, taxonomy-tree-edges 1408 ✓, aif-canonical-concepts 142 ✓, aif-owl-attack-edges 60 ✓
  • Investigation doc docs/investigations/2026-07-19-aif-attack-graph-export.md + export README both additive

0 secret / 0 PII / 0 machine-path leak across the 6 export+script+doc files.

△ Nit (non-blocking, wording) — body/PR title says "read-only, 0 write"; firsthand the script does write — to the export OUT_DIR (by design). The intent ("0 write to source taxonomy data") is accurate and verified (source CSVs untouched); suggest rewording "0 source-mutation, export-write by design" to avoid the literal contradiction for future readers.

Verdict: mergeable — delivers a sound typed-bipartite AIF attack-edge export, derived read-only from canonical sources, axiom-conformant 145/145, Reed-Norman distribution exact. The "0 write" prose is a wording nit, not a defect.

@jsboige
jsboige merged commit 900cee2 into master Jul 19, 2026
3 checks passed
@jsboige
jsboige deleted the docs/aif-attack-graph-export branch July 19, 2026 11:13
@jsboige jsboige mentioned this pull request Jul 19, 2026
Merged
jsboige pushed a commit that referenced this pull request Jul 19, 2026
Your @claude
docs(ontology): AIF export — add cross-family coherence note (Fallaci…
bd61351 
…es↔Virtues vocabularies disjoint)

Empirical finding (read-only, no fabrication): Fallacies (#828) and Virtues
(#829) reference DISJOINT AIF vocabularies (intersection = 0).
- Fallacies: 60 concepts, "<Topic>_<Type>" suffix (_Inference/_Conflict/_Scheme)
- Virtues: 14 concepts, "Argument from <Topic>" readable form

Both conventions co-exist as declared classes in the source AIF ontology
(Ontology/Resources/AIF.owl: 125 _Type forms + 15 "Argument from X" forms).
A lexical alignment is inferrable in several cases but NO materialised mapping
table exists in the source data. The export therefore does NOT unify the two
vocabularies — documented to prevent the false assumption that the two concept
spaces are pre-aligned. A consumer needing a unified AIF space must build that
mapping explicitly (modelling decision, not export).

Same lane (Cards/AIF), post-tag-safe, 0 write to sources. Amend to PR #829.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jsboige pushed a commit that referenced this pull request Jul 19, 2026
Your @claude
docs(ontology): link aif-export/ CSVs from ontology README (discovera…
f85198f 
…bility)

The ontology README (§2 Layer 3) documents the AIF attack-typed counts
(Fallacies 145, Virtues 222) but did not reference the consumable CSV exports
in aif-export/ added by #828 + #829. Add a short pointer in §3 so downstream
consumers (CoursIA ICT #7289, Layer C #790) can discover the bipartite
attack-edges, good-tenor edges, canonical concepts, taxonomy tree, and
inter-fallacy relations CSVs + the reproducible export scripts.

Same lane (Cards/AIF docs), post-tag-safe, 0 write to sources. Amend to PR #829.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jsboige added a commit that referenced this pull request Jul 21, 2026
@jsboige Your @claude
docs(ontology): AIF Virtues export — companion to #828 (double V-A/V-…
f8b02c6 
…B modelling, read-only) (#829)

* docs(ontology): AIF Virtues export — companion to #828 (double V-A/V-B modelling, read-only)

Extends the AIF export (#828, Fallacies) to the Virtues Taxonomy. Companion to
aif-attack-graph-export.py. Read-only on sources, post-tag-safe.

Honest verdict: the 222 virtues carry a DOUBLE AIF modelling of the same
population — two complementary views, never merged:
- V-A (CSV): 222 bipartite attack-edges (virtue -> node-TYPE RA/I/CA), 0 axiom
  violations (undercut->RA 206, undermine->I 13, rebut->CA 3).
- V-B (OWL aif#goodTenorOf): 222 virtue -> canonical argument-scheme edges
  (14 schemes: Rule 50, Commitment 40, Bias 27, Sign 26, ...).
Same 14-scheme vocabulary, different predicates. Not redundant, not contradictory.

Contrast with Fallacies (#828): Fallacies carry ONLY the attack-graph (V-A), no
goodTenorOf. Virtues are the dual (BOTH views). Do NOT fuse Fallacies-attacks +
Virtues-attacks — Virtues-attacks encode counter-arguments to fallacies.

Artefacts (docs/ontology/aif-export/):
- aif-virtues-attack-edges.csv       222 V-A bipartite attack-edges (PK-keyed)
- aif-virtues-good-tenor.csv         222 V-B virtue->scheme edges
- aif-virtues-canonical-concepts.csv 14 AIF schemes (skos refs)
- aif-virtues-schemes.csv            14 schemes distribution
- README.md                          + Virtues companion section

tools/aif-virtues-export.py: reproducible (Python stdlib, ~1s, 0 write to sources).

Refs: #828 (Fallacies AIF export), CoursIA ICT #7289, Layer C #790.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(ontology): AIF export — add cross-family coherence note (Fallacies↔Virtues vocabularies disjoint)

Empirical finding (read-only, no fabrication): Fallacies (#828) and Virtues
(#829) reference DISJOINT AIF vocabularies (intersection = 0).
- Fallacies: 60 concepts, "<Topic>_<Type>" suffix (_Inference/_Conflict/_Scheme)
- Virtues: 14 concepts, "Argument from <Topic>" readable form

Both conventions co-exist as declared classes in the source AIF ontology
(Ontology/Resources/AIF.owl: 125 _Type forms + 15 "Argument from X" forms).
A lexical alignment is inferrable in several cases but NO materialised mapping
table exists in the source data. The export therefore does NOT unify the two
vocabularies — documented to prevent the false assumption that the two concept
spaces are pre-aligned. A consumer needing a unified AIF space must build that
mapping explicitly (modelling decision, not export).

Same lane (Cards/AIF), post-tag-safe, 0 write to sources. Amend to PR #829.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(ontology): link aif-export/ CSVs from ontology README (discoverability)

The ontology README (§2 Layer 3) documents the AIF attack-typed counts
(Fallacies 145, Virtues 222) but did not reference the consumable CSV exports
in aif-export/ added by #828 + #829. Add a short pointer in §3 so downstream
consumers (CoursIA ICT #7289, Layer C #790) can discover the bipartite
attack-edges, good-tenor edges, canonical concepts, taxonomy tree, and
inter-fallacy relations CSVs + the reproducible export scripts.

Same lane (Cards/AIF docs), post-tag-safe, 0 write to sources. Amend to PR #829.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(ontology): reconcile aif-relations-graph count (1734 distinct vs README 1985 raw)

Resolves the crossLink delta tracked at tick 49 (README §2 Layer 2 = 1985 vs
aif-relations-graph.csv = 1734). Read-only investigation finding — NOT a bug,
a counting-methodology difference:

- README 1985 = RAW emitted assertions (pre-dedup).
- Raw OWL parse = 1977 (= README 1985 minus 8 schema self-definitions, one
  per verb across the 8 relation verbs).
- 1977 raw - 1734 distinct = 243 EXACT-duplicate triples (OWLSharp serializer
  idempotency quirk, not a data issue). The dedup is correct for a consumer.
- Symmetric verbs (mirrors/isRelatedTo/inverts/opposes) are emitted
  bidirectionally as two distinct triples (A,p,B)+(B,p,A) — both kept (not
  duplicates).

Added a count-reconciliation note to the aif-relations-graph.csv section of
the export README so consumers understand the 1734 vs 1985 difference.

Same lane (Cards/AIF docs), post-tag-safe, 0 write to sources. Amend to PR #829.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Your <your.email@example.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@jsboige jsboige mentioned this pull request Jul 21, 2026
Merged
jsboige added a commit that referenced this pull request Jul 21, 2026
@jsboige @claude
docs(ontology): clarify rebut→CA is export's mode-marker, not canonic…
0bf8436 
…al AIF target (#833 review)

Addresses NanoClaw's structural-review nit on the #7289 Phase-B consumption
contract: the new "Typed edges" bullet said `rebut` "attacks the conclusion,
CA-node", conflating the canonical semantic target (an I-node) with the
export's node-type marker (`CA-node`).

Verified against aif-attack-edges.csv: the export genuinely emits
`rebut → CA-node` (5 rows, node_type_axiom_ok=1), consistent with the
already-merged verdict table (#828). Reworded the bullet to state that `CA`
here is the export's *conflicting-application* mode-marker (keeping the three
attack types on distinct node-types = separable spectral channels), NOT the
canonical AIF target — in canonical AIF/ASPIC+ a rebuttal targets the
conclusion I-node. Consumer should read it as a type channel, not a node
instance.

Additive docs-only clarification; CSVs unchanged, 0 data change.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jsboige added a commit that referenced this pull request Jul 21, 2026
@jsboige Your @claude
docs(ontology): AIF export — directed/typed spectral consumption cont…
b442c65 
…ract (#7289 Phase-B) (#833)

* docs(ontology): AIF export — directed/typed spectral consumption contract (#7289 Phase-B)

Bridge note (tertiary deliverable of ai-01 dispatch #497/#498/bridge
deep-queue) for CoursIA ICT #7289 strate-6 Phase-B consumers, and the
nit raised in notebook #7341 review: the AIF attack graph is DIRECTED
(attacker -> abstract node-TYPE, bipartite) and TYPED
(undercut/undermine/rebut). A spectral/Laplacian consumer must preserve
both, or collapse the mode-of-attack signal the analysis wants to surface.

- Directed, not symmetric: standard L_sym on an undirected projection
  treats attacker and attacked node-type as interchangeable (they are not;
  direction is the semantics).
- Typed edges carry distinct modes: do not binarise undercut/undermine/rebut
  (RA-node / I-node / CA-node = three argumentative operations).
- Complementary graphs have different symmetries: aif-relations-graph.csv
  (bidirectional mirrors/isRelatedTo) vs aif-attack-edges.csv (typed
  directed) -- pick the graph matching the question, do not merge.
- Recommended (consumer choice, not exported): directed / magnetic hermitian
  Laplacian, or type-weighted / multi-channel edges.

Consumption contract only -- CSVs unchanged, 0 data change, 0 fabrication.
Read-only docs lane (same as #828/#829), post-tag-safe. Verdict QA = ai-01.

Refs: dispatch ai-01 -> po-2024 (msg-20260721T174854-ucdwi7).

Co-Authored-By: Claude-Code <noreply@anthropic.com>

* docs(ontology): clarify rebut→CA is export's mode-marker, not canonical AIF target (#833 review)

Addresses NanoClaw's structural-review nit on the #7289 Phase-B consumption
contract: the new "Typed edges" bullet said `rebut` "attacks the conclusion,
CA-node", conflating the canonical semantic target (an I-node) with the
export's node-type marker (`CA-node`).

Verified against aif-attack-edges.csv: the export genuinely emits
`rebut → CA-node` (5 rows, node_type_axiom_ok=1), consistent with the
already-merged verdict table (#828). Reworded the bullet to state that `CA`
here is the export's *conflicting-application* mode-marker (keeping the three
attack types on distinct node-types = separable spectral channels), NOT the
canonical AIF target — in canonical AIF/ASPIC+ a rebuttal targets the
conclusion I-node. Consumer should read it as a type channel, not a node
instance.

Additive docs-only clarification; CSVs unchanged, 0 data change.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Your <your.email@example.com>
Co-authored-by: Claude-Code <noreply@anthropic.com>
This was referenced Jul 21, 2026
Open
Merged
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@clusterManager-Myia clusterManager-Myia clusterManager-Myia left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@jsboige @clusterManager-Myia