contradiction-detector

name	contradiction-detector
description	Detect contradictions when synthesizing or consolidating notes. Flags factual disagreements, date conflicts, and outdated info with [!contradiction] callouts. Trigger: When consolidating notes, merging knowledge, or user says "find contradictions", "check conflicts", "contradiction scan".
license	MIT
metadata	{"author":"gentleman-programming","version":"1.0","source":"AgriciDaniel/claude-obsidian"}

Contradiction Detector

Flag contradictions when synthesizing information from multiple notes. When two sources disagree, surface them explicitly so the human resolves them — never silently pick a winner.

---

Core Principle

Knowledge bases accumulate contradictions over time. A note from January says "use REST", a note from June says "migrate to GraphQL". Both are true in their context but contradict each other if read without dates. The Contradiction Detector surfaces these conflicts explicitly using [!contradiction] callouts so synthesis products are trustworthy.

---

Contradiction Types

1. Factual Disagreements

Two notes make incompatible claims about the same subject.

Detection:

• Identify shared entities or topics across notes (same tags, same wikilinks, same subject headings)
• Extract factual claims: definitions, specifications, quantities, statuses
• Compare claims about the same entity — if they differ, flag as contradiction

Examples:

• Note A: "The API rate limit is 100 req/s" vs Note B: "Rate limit: 50 req/s"
• Note A: "We use PostgreSQL" vs Note B: "Database: MySQL"

2. Date/Timeline Conflicts

Events or decisions placed at incompatible points in time.

Detection:

• Extract dates and temporal markers ("before X", "after migrating to Y", "in Q2 2025")
• Build a timeline of events per topic
• Flag when the same event has conflicting dates or when temporal ordering is impossible

Examples:

• Note A: "Migrated to v2 in March" vs Note B: "v2 migration planned for Q4"
• Note A: "Deprecated in 2024" vs Note B: "Added support in 2025" (for the same feature)

3. Outdated Information

A newer note supersedes an older one, but the older note is still referenced.

Detection:

• Compare notes about the same topic with different modification dates
• If the newer note explicitly contradicts or updates the older one, flag the older as potentially outdated
• Check if the older note has incoming links — if yes, those links may be spreading stale info

Examples:

• Note from 2024: "Auth uses JWT tokens" → Note from 2025: "Migrated auth to OAuth2"
• Note from January: "Team size: 5" → Note from June: "Team grew to 12"

4. Definition Drift

The same term is defined differently in different contexts.

Detection:

• Find notes that define the same concept (shared title words, aliases, or explicit "definition" sections)
• Compare definitions — if they differ in substance (not just phrasing), flag as definition drift

Examples:

• Glossary note: "SLA = 99.9% uptime" vs Architecture note: "Our SLA target is 99.95%"
• Style guide: "Components use PascalCase" vs Onboarding doc: "Name components with kebab-case"

---

Detection Process

Step 1: Scope Selection

Accept input as:

• A list of specific notes to compare
• A tag or folder to scan (e.g., all notes tagged #architecture)
• A topic query (e.g., "everything about authentication")
• Entire vault (expensive — warn the user)

Step 2: Claim Extraction

For each note in scope:

1. Parse the note content into sections
2. Extract explicit claims — statements of fact, decisions, specifications
3. Tag each claim with: source_note, section, date (from frontmatter or inline), confidence
4. Normalize claims for comparison (lowercase entities, resolve aliases)

Step 3: Cross-Reference

Compare claims across notes:

1. Group claims by topic/entity
2. For each group, identify pairs that conflict
3. Score conflict severity: definite (mutually exclusive), likely (different but possibly compatible), possible (ambiguous phrasing)

Step 4: Report Generation

Produce a structured contradiction report with [!contradiction] callouts.

---

Output Format

Inline Callouts (for note synthesis)

When producing a synthesized note, embed contradictions inline:

## Authentication

The system uses OAuth2 for authentication.

> [!contradiction] Conflicting sources
> - [[Auth Design Doc]] (2025-06): "OAuth2 with PKCE flow"
> - [[API Spec v1]] (2024-11): "JWT bearer tokens"
> - **Resolution needed**: Which is current? Check with the auth team.

Standalone Contradiction Report

When scanning a set of notes, produce a full report:

# Contradiction Report

> Generated: {DATE}
> Scope: {notes/tags/folder scanned}
> Notes Analyzed: {count}
> Contradictions Found: {count}

## Summary

| Severity | Count | Type |
|----------|-------|------|
| 🔴 Definite | {n} | Mutually exclusive claims |
| ⚠️ Likely | {n} | Different claims, probably conflicting |
| ℹ️ Possible | {n} | Ambiguous, needs human review |

## Contradictions

### 1. {Topic}: {Brief description}

**Severity**: 🔴 Definite
**Type**: Factual Disagreement

| Source | Claim | Date |
|--------|-------|------|
| [[Note A]] | "Rate limit is 100 req/s" | 2025-01-15 |
| [[Note B]] | "Rate limit is 50 req/s" | 2025-06-20 |

**Suggested Resolution**: Note B is newer — verify if the limit was changed. Update the older note or mark it as superseded.

---

### 2. {Topic}: {Brief description}
...

## Recommended Actions

1. {Highest priority contradiction to resolve}
2. {Second priority}
...

---

Resolution Strategies

When contradictions are found, suggest one of these resolution strategies:

Strategy	When to Use	Action
Supersede	Newer note clearly updates older	Mark older note as [outdated], link to newer
Merge	Both notes have partial truth	Create a single authoritative note
Contextualize	Both are true in different contexts	Add context markers (e.g., "In v1..." vs "In v2...")
Escalate	Cannot determine which is correct	Flag for human review, add [!contradiction] callout
Archive	Older note is fully obsolete	Move to archive folder, remove incoming links

---

Configuration

contradiction-detector:
  severity_threshold: likely  # possible | likely | definite — minimum severity to report
  scope: tags                 # notes | tags | folder | vault
  scope_value: "#architecture"
  include_outdated: true      # flag outdated info as contradictions
  date_weight: high           # how much to trust newer notes over older
  output: report              # report | inline | both

---

Integration with Other Skills

• obsidian-consolidation: Run contradiction detection DURING consolidation to prevent merging conflicting info
• wiki-lint: Duplicate entities from wiki-lint are prime candidates for contradiction scanning
• auto-wikilink: After resolving contradictions, auto-wikilink can connect the authoritative notes

---

Anti-Patterns

1. Auto-resolving contradictions — NEVER silently pick one version. Always surface the conflict for human decision.
2. Ignoring dates — Temporal context is critical. A claim from 2024 vs 2025 is likely an update, not an error.
3. Scanning the entire vault — Start with a focused scope (tag, folder, topic). Full vault scans are noisy and expensive.
4. Treating all disagreements as contradictions — Different perspectives on the same topic are not contradictions. Only flag when claims are mutually exclusive or factually incompatible.
5. One-shot analysis — Run contradiction detection as part of every consolidation, not as a one-time cleanup.

---

Critical Rules

1. NEVER auto-resolve contradictions — always surface them for human review
2. ALWAYS include source notes, dates, and specific quotes in contradiction reports
3. ALWAYS use [!contradiction] callout format for Obsidian compatibility
4. ALWAYS suggest a resolution strategy for each contradiction
5. ALWAYS consider temporal context — newer notes may intentionally supersede older ones
6. NEVER flag stylistic differences as contradictions — only factual, definitional, or temporal conflicts
7. ALWAYS group contradictions by topic for easier triage