causal-attribution

🔐 Deterministic AI liability attribution engine. MCP server that generates cryptographically-signed CausalCertificates for AI incident forensics. EU AI Act Art.12 compliant.

Counterfactual attribution for multi-agent LLM failures - find which message broke the system, not which agent.

Find the decisive error step in a failed agent run by counterfactual replay of patched prefixes

Know when an agent trace cannot justify a cause.

Code and frozen experiment artifacts for target-ancestral causal discovery and deconfounded mechanical attribution.

Agent Eval tool: score, audit & explain agent trajectories — metrics + causal failure attribution in pure Go.