root_cause¶

Root Cause Analyzer — structured root cause analysis for AI safety incidents.

Provides three complementary methods: 5 Whys iterative causal chains, Fishbone (Ishikawa) categorized cause analysis, and Fault Tree Boolean logic decomposition with minimal cut-set analysis.

Quick start¶

# Full analysis (all three methods)
python -m replication root-cause --incident "Agent escaped sandbox" --severity critical

# Single method
python -m replication root-cause -i "Kill switch failed" -m fault-tree

# HTML report
python -m replication root-cause -i "Drift detected" -s high -f html -o rca.html

# JSON for programmatic use
python -m replication root-cause -i "Unauthorized replication" -f json

Methods¶

Method	Flag	Description
5 Whys	`5whys`	Iterative "why?" chain from symptom to root cause
Fishbone	`fishbone`	Ishikawa diagram across 6 AI safety categories
Fault Tree	`fault-tree`	Boolean logic tree with minimal cut-set analysis
All	`all` (default)	Run all three methods

Fishbone categories¶

Policy — Safety policies, contracts, governance
Monitoring — Observability, alerting, detection
Architecture — System design, boundaries, isolation
Human — Operator decisions, training, procedures
Environment — Infrastructure, dependencies, external conditions
Data — Training data, inputs, information flow

Programmatic API¶

from replication.root_cause import full_analysis, format_html

result = full_analysis("Agent escaped sandbox", "critical")
html = format_html(result)