{"doctrine":"LLM reads and explains. Deterministic code decides. Human owns exceptions.","ibm_seven_skills":[{"n":1,"skill":"System design","shipped_in":"FastAPI orchestrator + hard LLM/deterministic/human boundary"},{"n":2,"skill":"Tool & contract design","shipped_in":"Pydantic v2 schemas with extra='forbid' on every payload; 5 typed adapter contracts"},{"n":3,"skill":"Retrieval engineering","shipped_in":"5 fixture-backed adapters retrieve Programme data; cert extraction with cached fallback"},{"n":4,"skill":"Reliability engineering","shipped_in":"LOCAL_MOCK_MODE default; cached extraction fallback; cached reasoning fallback; UI retry"},{"n":5,"skill":"Security & safety","shipped_in":"Untrusted document text (RSK-01); read-only LLM; XSS-escaped UI; PII gitignored"},{"n":6,"skill":"Evaluation & observability","shipped_in":"Historical benchmark (94.6% path-match on n=522); append-only audit; 11 tests"},{"n":7,"skill":"Product thinking","shipped_in":"Beneficiary vs officer surfaces; plain-language reasoning; exception bands; Why-this-plan drawer"}],"rubric_alignment":{"Agentic Decision Intelligence (25)":["system design","tool contracts","retrieval"],"Policy Compliance & Governance (25)":["system design","security","observability","product"],"Technical Excellence & Data Integration (20)":["system design","tool contracts","retrieval","reliability"],"Impact on Service Transformation (15)":["observability (benchmark)"],"Demo, Explainability & UX (15)":["reliability","product thinking"]},"honest_claims":{"path_match_accuracy_2025":"94.6% (n=522, held-out)","twenty_pct_compliance_update_plans":"100% by construction","premium_dev_median_aed":557,"months_dev_median":10,"manual_baseline":"~5 working days","draft_latency":"<1 second (measured)","do_not_claim":"exact reproduction of officer premium/months — officers apply discretion routed to a human"}}