argument-architecture

Files (10)

• SKILL.md
• argument-templates.md · references/
• fallacy-catalog.md · references/
• irac-worked-examples.md · references/
• toulmin-worked-examples.md · references/
• argument-map.md · templates/
• counter-argument-log.md · templates/
• irac-blank.md · templates/
• qa-checklist.md · templates/
• toulmin-blank.md · templates/

SKILL.md (first lines)

# Argument Architecture Skill

Every serious argument has bones. This skill is those bones.

Used for legal submissions, thesis chapters, technical proposals — anything
where the reader must be moved from disagreement or neutrality to agreement
through structured reasoning.

---

## Skill Tree

| Module | Domain |
|---|---|
| 1 | IRAC++ — the legal argument spine |
| 2 | Toulmin Model — general argumentation structure |
| 3 | Deductive / Inductive / Abductive chains |
| 4 | Syllogistic form and its traps |
| 5 | Argument mapping — visualizing structure before drafting |
| 6 | Counter-argument anticipation |
| 7 | The Persuasion Paragraph — micro-structure |
| 8 | Logical fallacy audit |
| 9 | Bridge sentences — the connective tissue |
| 10 | Argument QA — testing before submission |

---

## Module 1: IRAC++

Classical IRAC (Issue, Rule, Application, Conclusion) is the legal spine.
IRAC++ adds two elements that separate competent submissions from strong ones.

```
I — ISSUE          What is the precise question?
R — RULE           What is the governing law?
   R+ — SUB-RULE   What sub-rules, tests, or standards apply?
   R++ — AUTHORITY What case, statute, or regulation establishes each?
A — APPLICATION    How do the facts meet or fail each element?
   A+ — COUNTER    What is the strongest opposing application?
C — CONCLUSION     The answer, restated with force
```

### The Issue Statement Test

An issue statement fails if any of the following is true:
- It's a topic, not a question
- It's a yes/no question when the real issue is a "how" or "to what extent"
- It buries multiple sub-issues under one label
- It uses conclusory language ("whether the landlord acted in bad faith" —
  presumes the framing; better: "whether s.51 permits the use asserted")

### The Rule Statement Test

A rule statement fails if:
- It paraphrases without citation
- It cites without quoting the operative words
- It omits the test the court will actually apply
- It states the rule without stating its exceptions

### The Application Test

An application fails if:
- It restates facts without connecting them to rule elements
- It skips an element
- It doesn't address the counter-application

---

## Module 2: Toulmin Model

For non-legal argumentation (thesis chapters, technical proposals), Toulmin
is the general-purpose structure.

```
CLAIM       What you're arguing
GROUNDS     Evidence supporting the claim
WARRANT     The reasoning link connecting grounds to claim
BACKING     Support for the warrant itself
QUALIFIER   Under what conditions the claim holds
REBUTTAL    Where the claim would fail
```

### Why Toulmin Beats "Claim + Evidence"

Naive argument: "X, because Y."
Toulmin: "X (claim), because Y (grounds), and Y implies X because Z (warrant),
which we know from W (backing), except when V is present (rebuttal)."

The naive version hides the warrant. The Toulmin version exposes it — and
exposed warrants can be tested, defended, or corrected.

Most argument failures happen at the warrant, not the grounds. Toulmin
forces the warrant into view.

---

## Module 3: Deductive / Inductive / Abductive Chains

Every argument runs on one of three engines. Know which you're using.

### Deductive

Structure: If premises are true, conclusion MUST be true.
Use when: legal rule application, mathematical proof, logical derivation.
Failure mode: false premise, invalid inference form.

Example:
```
P1: A landlord ending tenancy under s.51 must occupy in good faith
P2: This landlord's affidavits show no intention to occupy
C:  Therefore this s.51 notice is invalid
```

### Inductive

Structure: If premises are true, conclusion is PROBABLY true.
Use when: pattern arguments, empirical claims, evidence-weight arguments.
Failure mode: unrepresentative sample, hidden variables.

From dmang69/bc-legal-ai, MIT.

More from dmang69/bc-legal-ai