Skip to content

Latest commit

 

History

History
91 lines (70 loc) · 5.22 KB

File metadata and controls

91 lines (70 loc) · 5.22 KB

A hospital MRF still declares the superseded 2.0.0 template, seven months into v3.0.0

A finding from the first published mrf-honest cohort, observed 2026-08-14. This describes one published file on one date. It is not a ranking of any hospital, not a statement about care, and not a legal compliance determination.

What was done

On 2026-08-13, mrf-honest discover retrieved the CMS-conventional cms-hpt.txt document from https://www.cedars-sinai.org/cms-hpt.txt, which names one machine-readable file:

https://www.cedars-sinai.org/content/dam/cedars-sinai/billing-insurance/documents/951644600_CEDARS-SINAI-MEDICAL-CENTER_standardcharges.json

On 2026-08-14 at 00:47:02 UTC, mrf-honest scorecard performed one identified, bounded GET of that URL (robots.txt checked first; the path is allowed), verified the decoded body by SHA-256, and streamed the complete file through the local inspector. The exact evidence is the committed assessment row in data/cohorts/2026-08-14.assessments.jsonl:

Evidence Value
Decoded size 883,973,507 bytes
Content SHA-256 8c188ab3b02967785256d5867cf760778581e5e849f0fd1141955c2a2dcd2dbf
Declared version 2.0.0
Declared last_updated_on 2025-11-26
Charge items streamed 162,611 (scan completed; zero parser problems)
Payer-rate entries 1,033,280 (572,912 dollar; 408,093 percentage; 75,572 algorithm)

The finding

The file's required version field declares CMS template 2.0.0. CMS's JSON documentation for the machine-readable file states that v3.0 — which carries the CY 2026 OPPS/ASC final rule's newly required data elements — is "effective January 1, 2026, with CMS enforcement beginning April 1, 2026." As retrieved on 2026-08-14, more than seven months after the effective date and four months after enforcement began, the published file still declares the superseded template version.

The inspector emits this as one ERROR-severity conformance finding, CMS_V3_VERSION_UNEXPECTED, whose catalog message is deliberately narrow: "The template version is not the v3.0.0 version implemented here." Under the published file-grade policy, one dimension with errors makes this file a C.

Two consequences follow mechanically, and both are recorded rather than inferred:

  1. The v3-only lakehouse refuses the file (unsupported hospital JSON template version: '2.0.0'), so unlike the other five cohort files it carries no executable-contract evidence. Its scorecard page states that absence instead of implying a pass.
  2. Every downstream consumer that implements the current CMS schema faces the same fork: reject the file, or maintain a legacy v2 path CMS's own timeline has retired.

What cuts the other way

Neutrality requires stating what the file gets right, and it is a lot:

  • The envelope carries every field the v3 inspector requires — attestation with attester name, type_2_npi, license information, hospital name and address — so the content is closer to v3 expectations than the version string suggests.
  • The 884 MB body streams to completion with zero JSON problems: no BOM, no malformed items.
  • last_updated_on (2025-11-26) is inside the required annual window relative to the observation date, and it predates the v3 effective date. A plausible benign reading is that the hospital updates this file on an annual cadence that has not yet crossed 2026-01-01. That reading does not change what a consumer retrieves today, which is the point of the finding.

This inspection is narrower than CMS validation: an emitted finding means the selected check observed exactly what it says, and the absence of other findings is not a certificate that the rest of the file is valid.

Sources

A related pattern worth one paragraph

Four of the six files in this cohort — both UC Health files, Stanford Health Care's main file, and UC Davis Medical Center's file — begin with a UTF-8 byte-order mark. RFC 8259 §8.1 says implementations must not add a BOM to networked JSON, and the harm is not theoretical: this project's phase-0 measurement recorded Python's standard json.load failing outright on such a file. The inspector tolerates the BOM and records it as an INFO observation (JSON_UTF8_BOM_PRESENT) that does not lower any grade; it is noted here because a consumer with a strict parser will hit it before they hit anything else.

Corrections

If any evidence above is wrong, the correction path is a GitHub issue on this repository; the registry entry, the assessment row, and this document will be corrected and the correction dated. Being corrected is preferable to being counted right.