Corpus Analysis

Corpus Analysis and Gap Detection

This skill provides methods for analyzing corpus coverage, detecting knowledge gaps, and generating prioritized research tasks.

Coverage Analysis Methods

1. Source Distribution Analysis

Analyze how research sources map to planned chapters/sections.

Questions to ask:

• Which chapters have the most/least source support?
• Are sources evenly distributed or clustered?
• Which topics have only one source (single point of failure)?

SurrealQL query:

-- Count sources per chapter
SELECT chapter, count() as source_count
FROM section->cites->source
GROUP BY chapter
ORDER BY source_count DESC;

2. Entity Coverage Analysis

Identify which characters, locations, concepts, events are well-documented vs underrepresented.

Questions to ask:

• Which entities appear in only one source?
• Which entities lack descriptive detail?
• Which relationships are missing supporting evidence?

SurrealQL queries:

-- Entity mention frequency
SELECT name, count(<-supports<-source) as source_count
FROM concept
ORDER BY source_count ASC;

-- Characters with sparse descriptions
SELECT name, description, count(<-appears_in<-section) as appearances
FROM character
WHERE length(description) < 100
ORDER BY appearances DESC;

-- Locations not yet introduced
SELECT name, description
FROM location
WHERE introduced = false;

3. Topic Coverage Analysis

Map topics from BRD against corpus to find coverage gaps.

Questions to ask:

• Which BRD topics have zero corpus representation?
• Which plot points lack factual grounding?
• For nonfiction: which thesis components lack evidence?

SurrealQL queries:

-- Topics mentioned in sources
SELECT name, count() as mentions
FROM concept<-related_to<-source
GROUP BY name
ORDER BY mentions DESC;

-- Timeline gaps (missing events)
SELECT * FROM event
ORDER BY sequence;

-- Uncited knowledge gaps
SELECT question, context, created_at
FROM knowledge_gap
WHERE resolved = false
ORDER BY created_at ASC;

4. Source Quality Analysis

Evaluate reliability distribution across the corpus.

Questions to ask:

• What percentage of sources are high reliability?
• Are critical claims supported by high-quality sources?
• Which topics rely primarily on low-reliability sources?

SurrealQL queries:

-- Source reliability distribution
SELECT reliability, count() as count
FROM source
GROUP BY reliability
ORDER BY reliability DESC;

-- Sources by type
SELECT source_type, count() as count
FROM source
GROUP BY source_type;

-- Low-reliability sources supporting key concepts
SELECT
  <-supports<-source.title as source_title,
  <-supports<-source.reliability as reliability,
  name as concept
FROM concept
WHERE <-supports<-source.reliability IN ['low', 'very low'];

Knowledge Gap Detection Patterns

Pattern 1: BRD Requirements Without Corpus Support

Method: Compare BRD sections to corpus entities and sources.

Steps:

1. Extract key requirements from each BRD section
2. Query corpus for matching concepts, characters, locations
3. Flag requirements with zero or low matches

Example:

• BRD mentions "wireless operator training protocols at Beaulieu 1942"
• Query: SELECT * FROM concept WHERE name CONTAINS 'wireless' OR name CONTAINS 'Beaulieu'
• If no results: FLAG as high-priority gap

Pattern 2: Shallow Coverage (Single Source)

Method: Identify topics mentioned in only one source.

Why it matters: Single-source claims are fragile and hard to verify.

SurrealQL:

-- Topics with only one supporting source
SELECT name, count(<-supports<-source) as source_count
FROM concept
WHERE count(<-supports<-source) = 1;

Pattern 3: Missing Relationships

Method: Check for expected but missing graph edges.

Examples:

• Character mentioned but no ->knows-> relationships
• Location exists but never ->located_in-> any section
• Event with no ->precedes-> or ->follows-> temporal links

SurrealQL:

-- Characters with no relationships
SELECT name FROM character
WHERE count(->knows->character) = 0;

-- Events with no temporal ordering
SELECT name FROM event
WHERE count(->precedes->event) = 0
AND count(->follows->event) = 0;

Pattern 4: Timeline Inconsistencies

Method: Detect chronological gaps or conflicts.

SurrealQL:

-- Events without dates
SELECT name, description
FROM event
WHERE date IS NONE;

-- Sequence gaps (e.g., 1, 2, 5, 6 — missing 3 and 4)
SELECT sequence FROM event
ORDER BY sequence;

Pattern 5: Uncited Sections

Method: Find written sections without source citations.

SurrealQL:

-- Sections with no citations
SELECT * FROM section
WHERE count(->cites->source) = 0;

Prioritization Framework

Use this framework to prioritize research tasks based on impact and urgency.

High Priority (Blocks Multiple Sections)

Criteria:

• Gap affects 3+ planned chapters/sections
• Core to the BRD thesis/premise
• Required for major plot point or key argument
• Timeline-critical (early chapters need it)

Examples:

• "SOE training protocols" (affects multiple training scenes)
• "Lyon resistance network structure" (entire middle section depends on it)
• "Protagonist's historical timeline" (affects chronological consistency)

Research task template:

Priority: HIGH
Blocking: [list chapter/section IDs]
Query: [specific research question]
Context: [why this is needed, what we already know]
Success criteria: [what would resolve this gap]

Medium Priority (Blocks One Section)

Criteria:

• Gap affects 1-2 sections
• Adds depth but isn't critical to plot/argument
• Can be worked around if research fails
• Later chapters (writing not imminent)

Examples:

• "Daily life details in Lyon 1943" (enriches setting but not critical)
• "German counter-intelligence methods" (adds realism to one scene)
• "Specific wireless equipment specs" (detail-level enhancement)

Research task template:

Priority: MEDIUM
Blocking: [section ID]
Query: [specific research question]
Fallback: [how to proceed if research fails]

Low Priority (Nice to Have)

Criteria:

• Doesn't block any section
• Enhances detail or authenticity
• Can be added in editing pass
• Background/contextual knowledge

Examples:

• "Period-accurate slang terms"
• "Weather patterns in occupied France"
• "Secondary character backstory details"

Research task template:

Priority: LOW
Enhancement for: [section or theme]
Query: [research question]

Research Task Generation Templates

Template 1: Factual Gap

**Task**: Research [specific topic]
**Priority**: [HIGH/MEDIUM/LOW]
**Blocks**: [chapter/section IDs]
**Context**:
- BRD requires: [what the BRD says]
- Corpus has: [what we currently know]
- Gap: [what's missing]

**Research questions**:
1. [Specific question 1]
2. [Specific question 2]

**Success criteria**:
- [ ] Found 2+ reliable sources on [topic]
- [ ] Extracted key facts: [list expected facts]
- [ ] Resolved knowledge_gap:[id]

**Search strategy**:
- Academic databases: [keywords]
- Primary sources: [archives, documents]
- Web search: [specific queries]

Template 2: Character/Entity Gap

**Task**: Research [character/entity name]
**Priority**: [HIGH/MEDIUM/LOW]
**Blocks**: [section IDs]
**Context**:
- Mentioned in: [where entity appears in BRD/outline]
- Current knowledge: [what corpus has]
- Needed: [missing details]

**Research questions**:
1. Background/history: [specifics]
2. Relationships: [who/what they connect to]
3. Timeline: [when they appear, key dates]

**Success criteria**:
- [ ] CREATE/UPDATE entity with full description
- [ ] Establish relationships via RELATE statements
- [ ] Add timeline anchors (dates, sequence)

**Sources to check**:
- [Specific books, archives, websites]

Template 3: Thematic/Conceptual Gap

**Task**: Research [theme/concept]
**Priority**: [HIGH/MEDIUM/LOW]
**Blocks**: [section IDs]
**Context**:
- BRD theme: [core theme/argument]
- Current support: [sources that touch on this]
- Gap: [missing evidence, examples, or depth]

**Research questions**:
1. [Theoretical/conceptual question]
2. [Evidence/example question]
3. [Counter-argument/complexity question]

**Success criteria**:
- [ ] Found diverse perspectives on [concept]
- [ ] Identified concrete examples/case studies
- [ ] Created concept entity with supporting sources

**Expected outcomes**:
- 3+ sources with varied reliability levels
- Clear link to BRD thesis

SurrealQL Queries for Gap Analysis

Comprehensive Coverage Report

-- Get overview of corpus completeness
LET $total_sources = (SELECT count() FROM source)[0].count;
LET $total_characters = (SELECT count() FROM character)[0].count;
LET $total_concepts = (SELECT count() FROM concept)[0].count;
LET $open_gaps = (SELECT count() FROM knowledge_gap WHERE resolved = false)[0].count;

RETURN {
  sources: $total_sources,
  characters: $total_characters,
  concepts: $total_concepts,
  open_gaps: $open_gaps,
  source_reliability: (SELECT reliability, count() as count FROM source GROUP BY reliability),
  chapters_with_citations: (SELECT chapter, count() as cites FROM section->cites->source GROUP BY chapter)
};

Gap Detection by Section

-- Find sections with weak source support
SELECT
  id,
  chapter,
  sequence,
  count(->cites->source) as citation_count,
  word_count
FROM section
WHERE count(->cites->source) < 2
ORDER BY chapter, sequence;

Entity Relationship Completeness

-- Characters without sufficient context
SELECT
  name,
  count(->knows->character) as relationships,
  count(<-appears_in<-section) as appearances,
  length(description) as desc_length
FROM character
WHERE count(->knows->character) = 0
   OR length(description) < 50
ORDER BY appearances DESC;

Example Gap Detection Workflow

1. Load BRD: Read BRD requirements for next chapter
2. Query corpus: Run coverage analysis queries
3. Identify gaps: Compare BRD needs vs corpus results
4. Prioritize: Apply HIGH/MEDIUM/LOW framework
5. Generate tasks: Use templates to create research tasks
6. Store gaps: CREATE knowledge_gap SET question=..., context=..., resolved=false
7. Report: Summarize findings with specific task IDs

Output Format

When performing corpus analysis, provide:

## Corpus Analysis Report

### Coverage Summary
- Total sources: [count]
- Source reliability: [high: X, medium: Y, low: Z]
- Entities extracted: [characters: X, locations: Y, events: Z]
- Open knowledge gaps: [count]

### Gaps by Priority

#### High Priority (Blocking)
1. [Gap description] — Blocks: [sections] — Research: [topic]
2. ...

#### Medium Priority
1. [Gap description] — Blocks: [sections] — Research: [topic]
2. ...

#### Low Priority
1. [Gap description] — Enhancement for: [context]
2. ...

### Recommended Next Action
[Research these high-priority gaps / Continue to plan-write / etc.]