shubchat/loab

**Technical Summary** Evaluates LLM agents on mortgage origination through a multi-component scoring rubric that requires both correct outcomes *and* compliant processes—tool calls in sequence, proper inter-agent handoffs, forbidden action avoidance, and evidence chains—across six templated scenarios spanning prime/near-prime/sub-prime borrower profiles and fraud detection. Built on agentic orchestration with policy-bound decision routing (DTI thresholds, credit score gates, KYC sequencing) against Australian mortgage regulation, designed for multi-step workflows including Processing Officer, Underwriter, Credit Manager, and Financial Crime escalation paths. Current benchmark suite (v0.1.0) spans origination tasks with servicing, collections, and compliance modules in active development.