ICg-CaST Documentation

ICg-CaST is a synthetic research framework for testing an AI-integrated carcinogenomics causal-state theory. The active Python package is icg_cast. The starter-kit logic has been migrated into the package under src/icg_cast.

User Guide

• Theory
  • Modeled Layers
  • Causal Interpretation
  • Counterfactual Checks
• Quickstart for Domain Scientists
  • Browser Path
  • Command-Line Path
  • Python Path
• Simulation
  • Data-Generating Process
  • Reproducibility
  • Main Outputs
  • Assumptions
• CLI
  • Simulate
  • Make Demo
  • Train
  • Evaluate
  • Graph
  • Bench
• Streamlit Browser App
  • Install
  • Run
• API
  • Simulation
  • Signatures
  • Graph
  • Modeling
  • Optional Data Sources
• Data Sources
  • Supported Stub Adapters
  • Provenance
  • Access and License Notes
• Public-data calibration prototype
  • What gets calibrated
  • Quickstart (Python API)
  • Quickstart (CLI)
  • Tiny end-to-end example
  • Acceptance criteria
  • Data governance reminder
• Validation
  • Predictive Metrics
  • Mechanistic Checks
  • Cross-Species Human Relevance
  • Simulator Sanity Checks
  • External Validation

Mechanistic Modeling

• Mechanism-Bottleneck Causal Networks (MB-CNet)
  • Architecture
  • Do-operations on bottleneck units
  • Survival outcomes
  • Acceptance criteria
  • CLI usage
• ICg-Bench
  • DGP variants
  • Four scored tasks
  • Scoring and the leaderboard
  • CLI
  • Example script
  • Submission
• Coefficient registry
  • Card schema
  • Evidence levels
  • Coefficient uncertainty
  • Python API
  • CLI
  • Coverage status
  • How to edit a coefficient

Governance

• Ethics and Limitations
  • Synthetic Scope
  • Claims to Avoid
  • Real Data
  • Bias and Validity Risks

Reference dictionaries live under materials/:

• data_dictionary.csv

• intervention_dictionary.csv

• chemical_archetypes.csv

• kcc_to_aop_mapping.csv

• dataset_manifest.csv

• provenance_template.json