Every figure we publish is derived from CMS cost report worksheet coordinates. A wrong coordinate produces a number that looks perfectly reasonable, so we reconcile against hospices whose audited figures are independently available and we publish the result — including the metrics that fail.
0 of 1 metrics currently pass reconciliation. Metrics that do not pass are labelled unverified everywhere they appear, including in the API response.
| Metric | Verdict | Passed | Rate |
|---|---|---|---|
| total_operating_expenses | MIXED | 47/155 | 30% |
314 figures across 277 hospital-years at 99 hospitals (50 in AL, 25 in TX, 20 in FL, 18 in CT), read off audited financial statements: state-filed audits for government hospital authorities, and continuing-disclosure filings on EMMA for bond-issuing non-profits. Every figure is transcribed by hand from a named document and stored with a citation to that document and page; a figure we could not source is left blank rather than estimated. Tolerance is 0.5% for money, exact for counts.
The sample is deliberately drawn from hospices that publish accounts at all — non-profits, which file an IRS Form 990, and government authorities, which are audited by their state. Proprietary hospices publish neither and cannot be checked this way, so nothing here demonstrates that a for-profit chain fills the same cost-report lines the same way.
We do not interpolate. We do not carry a figure forward from a prior year. A metric whose inputs are missing is returned as null, never as zero. Where a provider filed an amended cost report we use the latest filing and say so.