Data & downloads

Everything on this site comes from one monthly download of the CMS Provider Data Catalog nursing home theme. The files below are the originals, byte for byte. For display, each CSV is converted to Parquet; values are never edited, computed, or filtered.

Source batch

theme_nursing-homes_current.zip · 39.1 MB · processed 2026-09-12
SHA-256 b3611b7b57554fdab9b62448dfa68bc563c2eb4956fe1787aaeecfba1b98e118

Original files

• NH_Data_Dictionary.pdf
• manifest.json
• theme_nursing-homes_current.zip

Datasets in this batch

Datasets carry their own refresh dates. Files in one theme download can differ in vintage; the dates above are CMS's, not ours.

Batch history

Reading this ledger

This ledger records what this site served and when. It proves the mirror's history, not CMS's; CMS's own history lives in its monthly archive of month-end snapshots. Each entry records what CMS was serving at the moment of that build's fetch, so several entries carrying one checksum are one batch served across many builds, and an entry's date is when this site last asked, not when CMS last changed. The zip's checksum identifies the archive, not the data: CMS can regenerate the container around byte-identical files, so a changed zip checksum can mean a repackage rather than new data. Each file's own checksum identifies its data, and the per-dataset modified dates are the vintage to read. The chain is kept in two places, here and in the repository, and a build that can read neither stops rather than start a new chain over the old one. Each entry also records the counts this site derives from that batch beside the checksum of the file they came from, so a number that moves while its source has not is caught as a defect in this pipeline rather than reported as a change in CMS's data.

Machine-readable: ledger.json.

Method and corrections

The pipeline reads every column as text so leading zeros and codes survive, converts each CSV to Parquet for delivery, verifies row counts match exactly, and publishes the untouched originals above. Values are never edited, computed, or filtered, and every name gets the same layout by the same rules. If something here does not match the CMS source files, that is a bug: report it or email NursingHomeRecords.org@gmail.com and it will be fixed.

This site makes no third-party requests: the fonts, the query engine, and the engine's Parquet extension are all served from this domain, alongside the data. Every build loads the finished site in a browser, records every request it makes, and fails if any of them goes to another host. That check is why this sentence is here. It was here once before while it was untrue. The engine was fetching its Parquet reader from extensions.duckdb.org on every facility page, and the check at the time only searched the published files for known hostnames, which cannot see a URL that the engine assembles while it runs. Both the retraction and the return are dated in the decision log.

Processing notes

• ownership: 658 rows carry no owner name (blank, or the literal 'None', CMS's no-owner convention); they appear in facility tables but not on the Owners page