git:20260921.420751c to git:20261016.2df8ea1

8 added, 1 removed. Audit A to A.

# bnlearn Framework Documentation Agents

## Overview

This document provides agent scaffolding for the bnlearn framework documentation subsystem. The documentation is organized to capture the robust capabilities of the `bnlearn` Python package for use with Bayesian Networks and structural learning.

**Status**: ✅ Documentation

**Version**: 1.0

---

## Purpose

Documentation for the `bnlearn` Python package (https://github.com/erdogant/bnlearn) for causal discovery, parameter learning, inference, and sampling methods. It serves as a probabilistic graphical model abstraction layer for causal network generation, inference, and visualization.

- In the GNN pipeline, `bnlearn` renders **discrete categorical Bayesian networks** (optional A/B/C/D/E matrices; multi-modality and multi-factor supported) and is **executable**, not render-only: rendered scripts run at Step 12 via [`src/gnn/execute/bnlearn/`](../../src/gnn/execute/bnlearn/) with output under `BNLEARN_OUTPUT_DIR`. Scripts skip when the `bnlearn` extra is absent (`gnn.utils.runtime_safety.framework_availability`). Continuous linear-Gaussian models are unsupported (`supports_continuous=False` in `src/gnn/render/framework_registry.py`).

+ In GNN, the `bnlearn` generator compiles complete stationary finite A/B/D models into native exact fixed-action filters: rendered scripts run at Step 12 via [`src/gnn/execute/bnlearn/`](../../src/gnn/execute/bnlearn/) with output under `BNLEARN_OUTPUT_DIR`. Scripts skip when the `bnlearn` extra is absent (`gnn.utils.runtime_safety.framework_availability`). Continuous linear-Gaussian models are unsupported (`supports_continuous=False` in `src/gnn/render/framework_registry.py`).

## Documentation Organization

The bnlearn Framework documentation system is part of the broader GNN (Generalized Notation Notation) pipeline and ecosystem.

### Related Systems

- **Main Documentation**: [docs/README.md](../README.md)

- **GNN Documentation**: [gnn/AGENTS.md](../gnn/AGENTS.md)

- **Pipeline Reference**: [src/gnn/AGENTS.md](../../src/gnn/AGENTS.md)

- **bnlearn Official Documentation**: [erdogant.github.io/bnlearn](https://erdogant.github.io/bnlearn/)

## Subsystems

- **README.md**: Comprehensive guide, features, pipelines, and code examples for `bnlearn` integration.

**Last Updated**: 2026-04-14

## Quick Navigation

- **Get Started**: See [README.md](README.md)

- **Pipeline Overview**: [src/gnn/AGENTS.md](../../src/gnn/AGENTS.md)

- **Full Documentation Index**: [docs/README.md](../README.md)

## Documentation Standards

All documentation in this subsystem follows the professional standards established in the GNN documentation system:

- **Unified**: Directly implements structural, parametric, and causal workflows.

- **Streamlined**: Code snippets directly replicate primary package behaviors.

- **Configurable**: Exposes metrics and methods (e.g., HC, EX, bic, k2).

- **Functional**: End-to-end functionality including synthetic data and continuous discretization.

- **Logged Tested Documented**: Conforms to the standard repository-wide validation schema.

---

**Status**: ✅ Production Ready

**Compliance**: Professional documentation standards

**Maintenance**: Regular updates with new features

+ The source compiler retains authored and canonical CPTs, source timesteps,

+ normalization provenance and model/script identities. Independent native tests

+ check complete posterior sequences. C/E policy optimization, causal discovery

+ and component composition remain outside this filtering contract. Metadata-only

+ compatibility calls still emit a labelled synthetic demonstration. See the

+ [renderer contract](../../src/gnn/render/bnlearn/README.md).