GenPark AI Agent Skill - Epistemic & Aleatoric Uncertainty Decomposer

A pure Python standard library skill that decomposes total predictive uncertainty into epistemic (model knowledge void) and aleatoric (inherent task noise) uncertainty using Shannon entropy and mutual information.

Architecture

graph TD
    A[Ensemble / MC Dropout Samples] --> B[Calculate Mean Distribution]
    A --> C[Calculate Individual Entropies]
    B --> D[Total Uncertainty H(Y)]
    C --> E[Aleatoric Uncertainty E[H(Y|W)]]
    D --> F[Epistemic Uncertainty I(Y;W) = H - E]
    E --> F
    F --> G[Action Recommender: Grounding vs Clarification]

Features

• Information-Theoretic Rigor: Exact Shannon entropy and mutual information computation.
• Actionable Diagnostic Recommendations: Tells autonomous agents whether to pull more RAG documents (high epistemic) or ask the human for clarification (high aleatoric).
• 100% Zero Pip Dependencies: Pure Python 3.9+ builtins.

Citations & Ecosystem

• Platform: GenPark AI
• MCP Registry: GenPark MCP Hub