Ebook Analysis: Non-Fiction Knowledge Extraction

You analyze ebooks to extract knowledge with full citation traceability. This skill supports two complementary extraction modes:

• Concept Extraction - Extract ideas classified by abstraction (principle → tactic)

• Entity Extraction - Extract named things (studies, researchers, frameworks, anecdotes) that persist across books

Core Principle

Every extraction must be traceable to its exact source. Citation traceability is non-negotiable. Extract less with full provenance rather than more without it.

Two Extraction Modes

Mode 1: Concept Extraction

For extracting IDEAS organized by abstraction level.

Use when: Analyzing a book for transferable ideas, building a concept taxonomy, understanding how abstract principles relate to concrete tactics.

Output: JSON files (analysis.json, concepts.json)

Example: "Spaced repetition improves retention" is a MECHANISM at Layer 2.

Mode 2: Entity Extraction

For extracting NAMED THINGS that can be cross-referenced across books.

Use when: Building a knowledge base where the same study, researcher, or framework appears in multiple books. The goal is entity resolution—recognizing that "Hogarth's framework" in Range is the same as "kind/wicked environments" mentioned elsewhere.

Output: Markdown files in knowledge base structure

Example: "Kind vs Wicked Environments" is a FRAMEWORK by Robin Hogarth.

Choosing a Mode

If you want to... Use Mode

Understand a book's argument structure Concept Extraction

Build a reference library across books Entity Extraction

Create actionable takeaways Concept Extraction

Track what researchers say across sources Entity Extraction

Both Run both modes sequentially

Entity Extraction Mode (Detailed)

Entity Types

Type What It Captures Example

study Research findings, experiments, data Flynn Effect, Marshmallow Test

researcher People and their contributions Anders Ericsson, Robin Hogarth

framework Mental models, taxonomies, systems Kind vs Wicked, Desirable Difficulties

anecdote Stories used to illustrate points Tiger vs Roger, Challenger Disaster

concept Ideas that aren't frameworks Cognitive entrenchment, Match quality

Extended Entity Type Guidance

Some entities don't fit cleanly into the five types. Guidelines:

Entity Kind Use Type Rationale

Simulations/Games (Superstruct, EVOKE) anecdote Illustrative events, even if hypothetical

Institutions (IFTF, WEF) researcher Organizations contribute ideas like individuals

Historical events (Challenger disaster) anecdote Stories that illustrate principles

Hypothetical scenarios anecdote Future scenarios from books like Imaginable

Thought experiments framework If systematic; otherwise concept

When uncertain: Default to anecdote for narratives/events, concept for ideas, framework for systematic methods.

Author-as-Subject Pattern

When the book's author is also a significant entity (e.g., Jane McGonigal in Imaginable):

Create a researcher entity if:

• Author has notable prior work or institutional affiliation

• Author appears in Wikipedia or other reference sources

• Author's background/credentials are relevant to understanding the book

• Other books in your collection might reference them

Skip if:

• Author is primarily known only for this book

• No external sources to verify/enrich the entity

Template addition for author-subjects:

Note

This researcher is the author of [Book] in our collection. Their frameworks and concepts are documented separately.

Entity File Template

[Entity Name]

Type: study | researcher | framework | anecdote | concept Status: stub | partial | solid | authoritative Last Updated: YYYY-MM-DD Aliases: alias1, alias2, alias3

Summary

[2-3 sentence synthesized understanding]

Key Findings / What It Illustrates

1. [Claim or finding with source] — Source: [Book], Ch.[X]

2. [Another claim] — Source: [Book], Ch.[X]

Key Quotes

"Quotable text here."

"Another memorable quote."

Sources in Collection

Sources NOT in Collection

• [Book that would enrich this entity]

Related Entities

• Other Entity - Relationship description

Open Questions

• [What we don't yet know]

Knowledge Base Structure

/knowledge/ ├── _index.md # Master registry ├── _entities.json # Searchable index (generated) │ ├── nonfiction/ │ ├── _index.md # Domain index │ ├── _[book]-quotes.md # Book-specific quotes file │ ├── studies/ │ │ ├── flynn-effect.md │ │ └── chase-simon-chunking.md │ ├── researchers/ │ │ ├── hogarth-robin.md │ │ └── tetlock-philip.md │ ├── frameworks/ │ │ ├── kind-vs-wicked-environments.md │ │ └── desirable-difficulties.md │ ├── anecdotes/ │ │ ├── tiger-vs-roger.md │ │ └── challenger-disaster.md │ └── concepts/ │ ├── cognitive-entrenchment.md │ └── match-quality.md │ ├── cooking/ # Domain-specific structure │ ├── techniques/ │ ├── ingredients/ │ └── equipment/ │ └── technical/ ├── patterns/ └── technologies/

Quotes Extraction

Quotable quotes are a distinct extraction type. For each book, create a quotes file:

File: _[book-slug]-quotes.md

Structure:

Quotable Quotes from [Book Title]

Author: [Author] Last Updated: YYYY-MM-DD

On [Theme 1]

"Quote text here."

"Another quote on same theme."

On [Theme 2]

"Quote on different theme."

What makes a good quote:

• Memorable phrasing that captures a key insight

• Self-contained (understandable without context)

• Surprising or counterintuitive formulation

• Useful for presentations, writing, or reference

Entity Extraction Workflow

• Scan book - Read through identifying named studies, researchers, frameworks, illustrative stories

• Check existing entities - Use kb-resolve-entity.ts to see if entity already exists

• Create or update - New entity → create file; existing → add as source

• Add quotes - Extract memorable quotes to quotes file

• Cross-link - Add Related Entities sections

• Regenerate index - Run kb-generate-index.ts

Entity Extraction States (KB0-KB5)

State Symptoms Intervention

KB0 No knowledge base Create directory structure

KB1 Structure exists, no entities Begin extraction

KB2 Extracting from book Create entity files

KB3 Entities created, not linked Add Related Entities

KB4 Linked, no index Run kb-generate-index.ts

KB5 Complete for this book Proceed to next book

Cross-Book Synthesis Workflow

Triggered when: 2+ books have been extracted to the knowledge base.

Goals:

• Find entities that appear in multiple books

• Identify conceptual connections between books

• Surface contradictions or complementary perspectives

• Update entity files with multi-source synthesis

Process:

Entity overlap detection

Find entities with 2+ sources

grep -l "Sources in Collection" knowledge/nonfiction/**/.md |
xargs grep -l "| . | .* |" | head -20

Or manually review entities updated with new source.

Conceptual connection mapping

• Compare frameworks across books (e.g., Range's "wicked environments" ↔ Imaginable's "futures thinking")

• Identify shared researchers (e.g., Tetlock appears in both Range and Imaginable)

• Look for complementary themes (prediction failure → preparation despite uncertainty)

Synthesis documentation For entities appearing in 2+ books, update the Summary section:

Summary

[Synthesized understanding from BOTH sources, noting agreements and differences]

Cross-book insights Document thematic connections in context/insights/cross-book-{theme}.md :

Cross-Book Insight: [Theme]

Books Contributing

• Range (Epstein) - [perspective]
• Imaginable (McGonigal) - [perspective]

Synthesis

[How the books complement or contradict each other]

Concept Extraction Mode (Detailed)

Concept Types (Abstract → Concrete)

Type Definition Example

Principle Foundational truth or axiom "Communities form around shared identity"

Mechanism How something works "Reciprocity creates social bonds"

Pattern Recurring structure or approach "The community lifecycle pattern"

Strategy High-level approach to achieve goals "Build trust before asking for contribution"

Tactic Specific actionable technique "Send welcome emails within 24 hours"

Abstraction Layers

Layer Name Abstraction Example

0 Foundational Universal principles "Humans seek belonging"

1 Theoretical Domain-specific theory "Community requires shared purpose"

2 Strategic Approaches and frameworks "The funnel model of engagement"

3 Tactical Specific methods "Onboarding sequences"

4 Specific Concrete implementations "Use Discourse for forums"

Relationship Types

Relationship Meaning When to Use

INFLUENCES A affects B Causal or correlational connection

SUPPORTS A provides evidence for B Citation, example, validation

CONTRADICTS A conflicts with B Opposing claims

COMPOSED_OF A contains B Part-whole relationships

DERIVES_FROM A is derived from B Logical conclusions

Concept Extraction States (EA0-EA7)

State Symptoms Intervention

EA0 No input file Guide file preparation

EA1 Raw file, not parsed Run ea-parse.ts

EA2 Parsed, not extracted LLM extracts concepts

EA3 Extracted, not classified Assign types and layers

EA4 Classified, not annotated Add themes, relationships

EA5 Single book complete Export or proceed to synthesis

EA6 Multi-book ready Cross-book synthesis

EA7 Analysis complete Generate reports

Concept Extraction Workflow

• Parse - Run ea-parse.ts to chunk book with position tracking

• Extract - Present chunks to LLM for concept identification with exact quotes

• Classify - Assign type (principle→tactic) and layer (0-4)

• Annotate - Add themes and functional analysis

• Link - Connect related concepts

• Export - Generate analysis.json, concepts.json, report.md

Available Tools

Parsing Tools

ea-parse.ts

Parse ebook files into chunks with metadata and position tracking.

deno run --allow-read scripts/ea-parse.ts path/to/book.txt deno run --allow-read scripts/ea-parse.ts path/to/book.epub --format epub deno run --allow-read scripts/ea-parse.ts book.txt --chunk-size 1500 --overlap 150

Output: JSON with metadata, chapters (if detected), and chunks with positions.

Knowledge Base Tools

kb-generate-index.ts

Scan knowledge base and generate searchable entity index.

deno run --allow-read --allow-write scripts/kb-generate-index.ts /path/to/knowledge

Output: Creates _entities.json with all entities, aliases, and metadata.

kb-resolve-entity.ts

Search for existing entities before creating duplicates.

deno run --allow-read scripts/kb-resolve-entity.ts "Flynn Effect" deno run --allow-read scripts/kb-resolve-entity.ts "Hogarth" --threshold 0.5 deno run --allow-read scripts/kb-resolve-entity.ts "kind learning" --json

Options:

• --threshold <0-1>

• Minimum match score (default: 0.3)

• --limit <n>

• Maximum results (default: 5)

• --json

• Output as JSON

Validation Tools

ea-validate.ts

Validate analysis output for citation accuracy and schema completeness.

deno run --allow-read scripts/ea-validate.ts analysis.json --report

Anti-Patterns

The Extraction Flood

Pattern: Extracting every potentially interesting phrase. Fix: Ask "Would I cite this?" before extracting. Quality over quantity.

The Citation Black Hole

Pattern: Extracting without preserving exact quotes or positions. Fix: Always capture: exact quote, chapter reference, context.

The Duplicate Entity

Pattern: Creating new entity without checking if it exists. Fix: Always run kb-resolve-entity.ts first.

The Orphan Entity

Pattern: Entities without Related Entities links. Fix: Every entity should connect to at least 2 others.

The Quote-Free Entity

Pattern: Entity captures ideas but no memorable phrasing. Fix: Include Key Quotes section with author's exact words.

The Single-Book Silo

Pattern: Analyzing books without cross-referencing. Fix: After 2+ books, run synthesis to find connections.

Example Workflows

Full Entity Extraction (Range Example)

1. Scan book chapter by chapter
2. Identify all named studies, researchers, frameworks, anecdotes
3. Create inventory document listing all potential entities
4. For each entity: a. kb-resolve-entity.ts "[entity name]" to check existence b. Create markdown file in appropriate type directory c. Fill in template with findings and citations d. Add Key Quotes section
5. Create _range-quotes.md with all memorable quotes
6. Update _index.md with new entities
7. kb-generate-index.ts to rebuild _entities.json

Quick Concept Scan

1. ea-parse.ts book.txt --chunk-size 2000
2. For each chunk, extract top 3-5 concepts
3. Classify by type and layer
4. Generate concepts.json and report.md

Output Persistence

Entity Extraction Output

File Location

Entity files knowledge/{domain}/{type}/{entity-slug}.md

Quotes file knowledge/{domain}/_[book]-quotes.md

Entity index knowledge/_entities.json

Domain index knowledge/{domain}/_index.md

Concept Extraction Output

File Location

Full analysis ebook-analysis/{author}-{title}/analysis.json

Concepts only ebook-analysis/{author}-{title}/concepts.json

Citations ebook-analysis/{author}-{title}/citations.json

Report ebook-analysis/{author}-{title}/report.md

Verification (Oracle)

What This Skill Can Verify

• Citation positions exist - Validate quoted text appears at claimed position

• Schema completeness - Required fields present

• Cross-reference integrity - Referenced entities exist

• Duplicate detection - Entity doesn't already exist (via kb-resolve-entity.ts)

What Requires Human Judgment

• Significance - Is this worth extracting?

• Classification - Is this really a "framework" vs "concept"?

• Relationship validity - Does A really influence B?

• Quote quality - Is this actually memorable?

Integration Graph

Inbound (From Other Skills)

Source Leads to

research Multi-book synthesis ready

reverse-outliner Structural data for concept extraction

Outbound (To Other Skills)

From State Leads to

Entity extraction complete dna-extraction (deep functional analysis)

Concept extraction complete media-meta-analysis (cross-source synthesis)

Complementary Skills

Skill Relationship

dna-extraction 6-axis functional analysis for annotation

reverse-outliner Structural approach for fiction

voice-analysis Author style fingerprinting

context-network Knowledge base maintenance

Calibration Data (from Range + Imaginable extractions)

By Book Density

Book Type Expected Entities Estimated Effort

Dense non-fiction (Range, Thinking Fast & Slow) 60-100 4-6 hours

Moderate non-fiction (most business books) 30-50 2-3 hours

Light non-fiction (popular science) 15-30 1-2 hours

Technical books 20-40 2-3 hours

By Book Subtype

Different non-fiction subtypes yield different entity profiles:

Subtype Example Entity Profile Expected Count

Research synthesis Range Many studies, researchers, frameworks 60-100

Methodological/How-to Imaginable Many frameworks, few studies 30-50

Memoir/Narrative Educated Few frameworks, many anecdotes 20-40

Reference Technical manuals Many concepts, few anecdotes Variable

Research synthesis books cite many studies and researchers, connecting ideas across domains. Methodological books teach techniques and frameworks but cite fewer external sources. Memoir/narrative books use personal stories to illustrate points rather than research.

Metadata Reliability Warning

Book classification metadata (Calibre tags, library categories) is often:

• Wrong - Fiction/non-fiction misclassified

• Generic - "General Fiction" or "Self-Help" applied broadly

• Inconsistent - Same book categorized differently across sources

Always verify classification makes sense before extraction. A "fiction" tag on a methodology book like Imaginable is a metadata error.

Reasoning Requirements

Standard Reasoning

• Single chunk concept extraction

• Type/layer classification

• Simple relationship identification

• Individual entity creation

Extended Reasoning (ultrathink)

Use extended thinking for:

• Multi-book synthesis - requires holding multiple networks simultaneously

• Contradiction detection - semantic comparison across sources

• Theme emergence - identifying patterns across large sets

• Knowledge gap identification - reasoning about what's missing

Trigger phrases: "synthesize across books", "find contradictions", "identify gaps", "comprehensive analysis"

What You Do NOT Do

• Extract without citation traceability

• Create entities without checking for duplicates

• Skip the linking phase (orphan entities are not useful)

• Leave entities without quotes

• Treat fiction as non-fiction

• Use regex for semantic analysis (LLM judgment only)