SHShimazuSystems·Jun 5, 2025
TSUKUYOMI - Modular Intelligence Analysis Framework (40+ modules for structured analysis)
I've been working on TSUKUYOMI, a modular framework that implements intelligence community analytical standards in AI systems. It's essentially a collection of 40+ specialized modules that can be orchestrated to perform structured analysis on complex data.
GitHub: https://github.com/ShimazuSystems/TSUKUYOMI/
What it does
Instead of relying on black-box AI responses, TSUKUYOMI enforces structured analytical methodologies:
- Analysis of Competing Hypotheses (ACH) - Systematically evaluates multiple explanations
- Source reliability scoring - A-F reliability, 1-6 credibility (IC standards)
- Multi-source correlation - Temporal, spatial, semantic, network analysis
- Confidence quantification - Bayesian fusion with uncertainty propagation
Technical approa
Read the rest
ch
modules/
├── data_recognition_ingestion.tsukuyomi # Multi-INT data processing
├── correlation_analysis.tsukuyomi # Cross-source validation
├── strategic_scenario_modeling.tsukuyomi # Probabilistic forecasting
├── flight_data_analysis.tsukuyomi # ADS-B/MLAT processing
└── web_search_osint.tsukuyomi # OSINT collection
Each module follows a structured execution sequence with explicit reasoning chains. The orchestration core dynamically selects and chains modules based on requirements.
Example: OSINT Analysis
{
"execution_sequence": [
{
"operation": "Source Reliability Assessment",
"actions": ["evaluate_source_access", "check_corroboration", "assign_confidence"],
"output": "//SOURCE: Reliability B / Credibility 2"
},
{
"operation": "Entity Extraction",
"actions": ["identify_persons", "map_relationships", "extract_technical_indicators"],
"output": "//RESULT: 15 entities identified with confidence scores"
}
]
}
Why I built this
Current AI systems are great at pattern matching but poor at structured reasoning. In intelligence analysis, you need:
- Explicit evidence chains
- Alternative hypothesis consideration
- Quantified uncertainty
- Source attribution
- Audit trails
TSUKUYOMI provides these through modular components that enforce analytical rigor.
Current implementation
- Runs in Claude AI environment (upload the .tsukuyomi files)
- Modules are JSON-based configurations defining analytical workflows
- Includes specialized processors for HUMINT, SIGINT, GEOINT, OSINT, CYBINT
- Outputs professional intelligence products with proper confidence levels
Interesting technical bits
1. Multi-source information fusion (MSIF) at three levels: data, feature, and decision
2. Bayesian confidence integration across heterogeneous sources
3. Dynamic module orchestration based on information requirements
4. Classification-aware processing (handles UNCLASSIFIED through TOP SECRET)
Use cases beyond intelligence
- Research: Systematic literature review with source credibility scoring
- Business: Competitive analysis with structured methodology
- Engineering: Failure analysis using ACH framework
- Security: Threat correlation across multiple data sources
Future development
- Standalone Python implementation
- REST API for integration
- Custom module development SDK
- Visualization components for link analysis
Looking for feedback on:
1. The module architecture approach
2. Other domains where structured analysis would be valuable
3. Integration ideas with existing analytical workflows
The code is MIT licensed. Contributions welcome, especially domain-specific modules.