A Power BI report on CMS Medicaid State Drug Utilization Data — spend concentration, brand-to-generic substitution opportunity, per-capita state outliers, and a 12-month ensemble forecast. 5 report pages, 17 interactive visuals, refreshable on demand.
How this was built — pipeline + 16-evaluator LLM audit harness
Spec → bronze → silver → forecast → TOM model → 3-tier scalable-oversight pass
Expand
Healthcare Dashboard Ops
Power BI evaluation harness · scalable-oversight pattern
16 evaluators across three tiers run against a generated Power BI dataset and emit a severity-banded AUDIT.md plus a weighted composite verdict. The canonical run uses CMS Medicaid State Drug Utilization Data — 31,028,144 rows across 6 years.
Portfolio demonstration · public data only
This page is a portfolio demonstration of a Power BI evaluation harness applied to the CMS Medicaid State Drug Utilization Data — a publicly-available dataset published by the Centers for Medicare & Medicaid Services. No protected health information, beneficiary identifiers, or claim-level records are included. SDUD is state-level aggregated by drug and quarter.
Forecasts shown are illustrative model output, not policy guidance. Figures are for software-design demonstration purposes and should not be used for clinical, financial, or regulatory decision-making.
dashboard_spec.yml drives every downstream phase.
Bronze rows
31,028,144
DAX measures
39
Pages × visuals
5 × 17
Evaluators
16
16 evaluators · click for rubric detail
Reuse
Drop in a new dashboard_spec.yml with a different domain and data sources — every judge except domain_relevance runs unchanged. The full harness, DAX templates, judge code, and the canonical Medicaid run are on GitHub.