Methodology

Each source has a registered adapter. A harvest downloads the full official release (for CMS, the monthly CSV distribution from the Provider Data Catalog), archives it with a sha256 checksum, and records every row in a per-record ledger with first-seen, last-seen, last-changed and removed dates. Raw values are never overwritten.

The build normalizes raw rows into canonical entities and relationships, resolves identity, computes rollups, writes reverse indexes for graph navigation, and publishes versioned JSON shards with a manifest of counts, coverage, freshness and checksums. A validation stage checks referential integrity, counts against the raw store, universe rules and a reference facility before anything is published.

Facilities are identified by CMS Certification Number (CCN). We do not merge records because names or addresses look alike: separately certified providers at the same normalized address go to a review queue marked Unverified and stay separate.

Operators are the CMS legal business name of each certified provider, grouped only on an exact match after normalizing case, punctuation and corporate suffixes Corroborated. Chains use the CMS-assigned chain ID Verified. Owner organizations from the CMS Ownership file are linked across facilities by exact normalized name, because CMS publishes no owner identifier. Individuals are shown per facility and never merged across facilities by name.

Operator, chain, direct owner, indirect owner, managing-control entity, officer, lender and additional disclosable party are distinct roles and are never flattened into one entity.

Counties are assigned by matching the CMS county name to the Census county list; where the name is missing or ambiguous we locate the CMS coordinates in Census county boundaries (point in polygon), and mark that assignment Inferred.

Every citation keeps its source-native identity: CMS prefix and tag (F = health, K = Life Safety Code, E = emergency preparedness), CMS category, scope/severity letter and correction disposition. A citation is one row per CCN × survey date × tag.

Normalized concepts (clinical care, resident rights & abuse/neglect, medication, infection prevention, staffing, food & nutrition, environment & life safety, emergency preparedness, administration) come from an explicit, published category-level crosswalk Inferred. When state codes are added they will be crosswalked with evidence and confidence — never equated because their descriptions sound similar.

Scope and severity follow the CMS grid: A–C no actual harm with potential for minimal harm; D–F potential for more than minimal harm; G–I actual harm; J–L immediate jeopardy. Scope is isolated, pattern or widespread. “Corrected” does not erase a citation; we show both the finding and its disposition.

CMS publishes health citations from roughly the three most recent standard surveys plus complaint and infection-control inspections from the prior three years; penalties cover about three years. Older findings drop out of the files. Consequently a facility with no citations in the window has no citations in the window — not a clean history. Counts across states are not directly comparable without accounting for survey practice and timing.

CMS Five-Star ratings, PBJ staffing and quality measures describe CMS-certified nursing homes only. They are never shown for state-licensed assisted living. Every page and aggregate states the universe it covers, and no CMS count is presented as the number of all senior-care facilities in the US. “Beds” means CMS-certified beds unless labeled otherwise; licensed beds, units and reported capacity are kept as separate fields.

Verified

Stated directly by the official source record.

Corroborated

Linked by an exact identifier or exact normalized legal name.

Inferred

Derived by our method (e.g. concept grouping), not stated by the source.

Unverified

A possible link we have not confirmed. Never merged.

Historical

True as of a past date. Not current status.

Not found

We looked and found no record. This is not evidence that no issue exists.

Not public

The regulator does not publish this as open data.

No resident-level data and no protected health information. Proximity to environmental sites or flood zones is context, not evidence of exposure or risk. Ratings and measures are not guarantees of any individual resident’s experience. See corrections to dispute a record.