For the complete documentation index, see llms.txt. This page is also available as Markdown.

Caveats & Coverage

This dataset is built to be cited in diligence, underwriting, and compliance work, which means being precise about what it can and cannot see. Every caveat below is a property of the federal source data, stated so you can reason about it — not a disclaimer. Read this page before drawing a conclusion from an ownership interval, a staffing measure, or a closure signal.

Data coverage

Effective-dated ownership intervals: Feb 2023–present (PECOS). Labeled owner-name observations: Jan 2019–present (Care Compare).

This distinction matters for any ownership analysis: intervals and name observations are not interchangeable. DATA_DICTIONARY, and the flag methodology. The ID-keyed, effective-dated ownership intervals in OWNERSHIP_HISTORY (the ones that carry interval math, is_current, and every ID-based count) begin in February 2023, when CMS's PECOS ownership data becomes usable as dated intervals. Ownership history before that date ships only as labeled owner-name observations from the Care Compare ownership feed, back to January 2019 — carried as name-grain rows (source = 'care_compare_name'), never as intervals, and never counted in any ID-based figure. When you see an ownership interval, it is a PECOS interval from Feb 2023 onward. When you see a pre-2023 owner, it is a labeled observation, not a dated interval.

Ownership caveats

  • Self-reported ownership is a floor, not a census. PECOS ownership is what owners disclosed to CMS. Absence of a disclosed owner is a gap in the federal record — not proof that no such owner exists. Transactions structured to avoid federal disclosure are invisible to this product, and we do not infer them.

  • PE / REIT / holding-company flags are CMS self-reported. A facility not flagged as private-equity-owned may still have PE involvement that was never disclosed. Absence of a flag is not absence of involvement.

  • Effective dates carry roughly ±1 month of uncertainty. valid_from_observed is the first monthly vintage in which an ownership relationship appeared — a publication date, not the date the transaction closed. valid_from_reported is the owner's self-reported association date, carried verbatim and never used as the effective date. The two can disagree, and date_confidence labels which case each row is.

  • Care Compare owner names are display data, not identity. Owner-name strings churn heavily month to month (largely re-keying, not real change), so identity is always the CMS-assigned ID. Care Compare name rows are bridged to an ID as a conservative hint and are never used to count facilities or detect change.

Staffing (PBJ) caveats

  • PBJ lags roughly two quarters. Payroll-Based Journal staffing is published quarterly and arrives about two quarters behind. The most recent months of FACILITY_MONTH will typically have no evaluable PBJ-based flag yet.

  • Suppressed or unreported staffing is NULL, never zero. A quarter that is partial or missing yields NULL staffing measures and a NULL staffing flag — not a fabricated zero. HPRD measures are NULL when no resident-days were reported.

  • No statutory minimum is applied. The staffing flag compares reported HPRD against CMS's own case-mix expectation. The rescinded April-2024 federal HPRD staffing minimum appears nowhere in this product — not as a threshold, a column, or a benchmark.

Facility identity caveats

  • A CCN is a certification identifier, not a building. The CMS Certification Number identifies a Medicare/Medicaid certification, which can be reassigned, terminated, or re-issued. A CCN is not a stable identifier for a physical facility over time. ccn_reuse_suspect and successor_ccn_hint in FACILITY_MASTER are observed-only heuristics that flag likely reuse — never name-based inference.

  • Termination is not physical closure. A certification termination in the Provider of Services file, or a disappearance from the monthly snapshots, is a change in the federal record. The building may still operate under a new CCN, a new owner, or a different program. Closure signals are observed events, not statements that a facility's doors have closed.

Chain caveats

  • Chains are CMS-reported only. chain_id and chain membership come directly from CMS's own chain designations in the provider-information file. There are no inferred chains and no name-based grouping. no_chain_reported is an explicit value (a reported blank), not a silent gap.

  • chain_id coverage begins where CMS's chain reporting begins. Chain-membership history only extends as far back as CMS populated the chain_id field; earlier months carry no chain interval.

  • Chain aggregates can be partial. CHAIN_MONTH rollups carry a coverage_ratio and an aggregates_partial flag (set when under 90% of members had the underlying measure). Read aggregates with their coverage ratio.

Snapshot gap caveats

CMS does not publish every feed every month, and each feed family has its own gaps. The product never does month arithmetic across a gap — it reads the previous available month from SNAPSHOT_CALENDAR — so a missing month can never manufacture a phantom event or a phantom disappearance.

  • Provider-information is missing 2026-01.

  • Care Compare ownership is missing every December from 2019 through 2024, plus 2025-01 and 2025-08.

During a gap month, disappearance signals are not emitted, and coverage-dependent flags (ownership-disclosure) are NULL rather than 0.

Closure-signal caveats

  • Observed-only, and held pending during the CHOW lag. Closure signals fire only on observed events (snapshot disappearance across adjacent months, POS termination, beds → 0). A signal within CMS's official change-of-ownership publication lag, with no official CHOW yet explaining it, is held at pending_confirmation — the deactivation flag is NULL, not 1, until the window passes.

  • Provider of Services SNF history is shallow. The POS feed is a corroborating source for termination dates only. SNF/NF rows appear only in recent POS releases (from mid-2025 onward) — earlier POS releases carried no nursing-home rows — so POS corroboration is not available for historical months.

What this dataset is not

  • It is not a prediction of closure, distress, or care quality. Every flag is a deterministic signal about what the federal record shows, with a published formula and a versioned threshold.

  • It is not complete ownership monitoring. It is a view over CMS-disclosed ownership; undisclosed structures are invisible and are flagged as dead ends, not resolved.

  • It is not a real-time feed. It reflects CMS's published snapshots, with the publication lags noted above.

Last updated