JFK Assassination Document Analysis System

This system uses the Anthropic Claude AI to analyze declassified JFK assassination documents, extract key information, identify connections between entities, and surface potentially significant findings.

Setup

Prerequisites

• Python 3.8+
• pip (Python package manager)
• JFK assassination documents in PDF format (place in jfk_pdfs/ directory)
• Anthropic API key

Installation

1. Clone this repository or download the files

2. Install required dependencies:

   pip install -r requirements.txt
   

3. Create a .env file in the project root with your Anthropic API key:

   ANTHROPIC_API_KEY=your-api-key-here
   

Getting an Anthropic API Key

1. Go to https://console.anthropic.com/
2. Sign up for an account or log in
3. Navigate to API Keys section
4. Create a new API key
5. Copy the key and add it to your .env file

Files and Usage

1. Initial Document Processing

File: parse_pdfs.py

This script processes PDF documents to extract meaningful information related to the JFK assassination.

python parse_pdfs.py

• Converts each PDF to images
• Analyzes each page using Claude
• Looks for evidence related to 10 key conspiracy categories
• Generates initial JSON analysis files and summaries
• Creates a global summary of findings

2. Enhanced Analysis of Initial Results

File: parse_responses.py

This script analyzes the initial findings to identify the most significant documents and connections.

# Basic usage - processes most recent output directory
python parse_responses.py

# Process a specific output directory
python parse_responses.py --output output_20250318_180803

# Filter by category
python parse_responses.py --category "WITNESS_TESTIMONIES"

# Filter by entity
python parse_responses.py --entity "Lee Harvey Oswald"

# Show only documents with connections between entities
python parse_responses.py --connections-only

# Adjust confidence threshold
python parse_responses.py --min-confidence 7

Outputs:

• significant_findings.txt: Detailed report of findings
• high_confidence_findings.csv: Tabular data for further analysis
• entity_relationships.csv: Mapping of entity relationships
• structured_findings.json: Complete data in JSON format
• knowledge_graph.json: Network visualization data

3. Deep Analysis of High-Confidence Findings

File: final_parse.py

This script uses Claude 3.7 Sonnet (the most powerful model) to perform deeper analysis on the most promising documents.

# Analyze top 5 most promising documents (default)
python final_parse.py

Outputs detailed JSON analyses and a comprehensive summary in a timestamped directory.

4. Targeted Document Analysis

File: analyze_document.py

This script analyzes a specific document by ID with Claude 3.7 Sonnet.

# Analyze a specific document, automatically finding the most relevant page
python analyze_document.py 104-10332-10023

# Analyze a specific page
python analyze_document.py 104-10332-10023 --page 5

# Analyze all relevant pages in a document
python analyze_document.py 104-10332-10023 --all-pages

Provides a focused analysis of a single document, useful for investigating specific leads.

Output Directories

• output_TIMESTAMP/: Contains results from parse_pdfs.py
• output_final_TIMESTAMP/: Contains results from final_parse.py
• doc_analysis_TIMESTAMP/: Contains results from analyze_document.py

Understanding the Analysis

The system evaluates documents based on:

1. Relevance: Connection to the JFK assassination
2. Credibility: Quality of the evidence
3. Entity Connections: Relationships between people and organizations
4. Contradictions: Inconsistencies with the official narrative
5. Significance: Historical importance of findings

Example Usage Workflow

1. Process all PDFs with initial analysis:

   python parse_pdfs.py
   

2. Identify significant documents and connections:

   python parse_responses.py
   

3. Run deeper analysis on promising documents:

   python final_parse.py
   

4. Investigate specific documents of interest:

   python analyze_document.py DOCUMENT_ID
   

Notes

• Processing large numbers of documents can be time and API-cost intensive
• The quality of analysis depends on document legibility
• The system uses a tiered approach, using cheaper models for initial screening and more powerful models for promising documents

Troubleshooting

• If you encounter JSON parsing errors, the scripts include fallback mechanisms
• For best results, ensure PDFs are high quality and text is legible
• If a script crashes, you can usually resume by running it again - it will skip already processed documents