Payer Claims Lakehouse
A member-level claims lakehouse on the CMS synthetic Medicare files. Enrollment and claims arrive in yearly batches, pass data contracts, land in versioned bronze, and are built into silver and gold with dbt on DuckDB. This page is generated from the gold marts and from the pipeline's own load log and test results.
Synthetic data. CMS generated these files with Synthea and warns they have very limited inferential value; payments and lengths of stay do not resemble real Medicare. The rates below show how the measures are built and tested. They are not findings about Medicare, hospitals, or members.
Each stay gets the first readmission exclusion that applies; the rest are index stays. Discharge status is always 1 (home) in this source, so the death, against-advice, and transfer exclusions never fire here. An end-to-end test injects those codes to exercise them.
Diagnoses and ICD procedures are checked against the ICD-10 code set of each claim's fiscal year; revenue-line HCPCS codes are checked for CPT Category I or HCPCS Level II format (no CPT descriptions are stored). Problems are measured, not quarantined, because they are systemic in this source.
Every arrival passes its data contract before it reaches bronze, and every model is built and tested by dbt build; an error-level test failure stops the run. Each mart is recomputed from silver in a reconciliation test.
Warnings are visibility checks, not gates. Each one flags a known property of this synthetic source.