Methodology — How We Measure

Summary

HomeCareIntent watches official state licensing records for home care providers. It keeps every version of each source file, compares versions to detect what changed, and turns each change into a dated event tied to an organization. Public pages show aggregates and a limited set of fields for Live jurisdictions; figures never include sample data.

Pipeline

01Official source. A state licensing file, open data portal or roster, recorded with its regulator and URL.
02Immutable snapshot. Each download is stored unchanged with its retrieval time. Snapshots are never edited or deleted, so every figure can be traced back to the file it came from.
03Validation. Automated quality checks run on every download. A failed check stops publication.
04Normalization. Records are brought into a consistent form, keeping the original values.
05Entity resolution. Records are matched to organizations by deterministic rules. Uncertain matches go to human review.
06Change detection. Each snapshot is compared with the previous successful one.
07Events. Every change becomes an append-only event with its source record, snapshot, detected date and effective date.
08Intent scoring. Deterministic rules score organizations from their recent events. Scores are interpretation and appear only in the product.

Sources

Primary sources are official state licensing records; the source registry lists the regulator, official source, source type and stated cadence for each state. Federal CMS files cover Medicare-certified home health agencies only and are never presented as non-medical home care coverage. Third-party enrichment, such as websites or contact fields, appears only in the product, labeled with provider and date, and is never used in public counts.

Freshness

Two schedules are always stated separately. Source publication cadence is how often the government publishes updates, quoted as the source states it, or “not stated by the source”. Check frequency is how often HomeCareIntent downloads the source. A change reaches public pages after the source publishes it and the next successful check completes. Government sources update daily to monthly, so no figure is instantaneous. If checks for a Live jurisdiction stop succeeding, it is marked Degraded and publishing pauses until they recover.

Normalization and entity resolution

Each state's license types are mapped to common categories (non-medical home care, personal care, home health and others), and only types relevant to home care are counted. License statuses are mapped to a common set, such as active, pending, suspended and closed, so states can be compared. Records are matched to organizations across sources and snapshots by deterministic rules; matches that are not certain go to human review instead of being merged automatically. No language model merges entities. Merges are audited and reversible.

Change detection, baseline and backfill

The first complete ingestion of a source is the baseline. It records what already exists and creates no new-agency events, with one exception: when the source publishes an official issue date inside the lookback window, a New license event is backfilled with that issue date as its effective date. These rows are labeled “effective date from official issue date”. History may begin with an archived copy of the same official file, ingested through the same connector and labeled as reconstructed history.

How events are dated

Source-dated events carry a date the source publishes, such as a license issue date. Window-dated events do not: the source shows only the new state, so the change happened somewhere between the previous snapshot and the one where HomeCareIntent saw it. Address, administrator and many status changes are window-dated.

A window-dated event counts toward a period, such as “events in the last 30 days” or a calendar month, only when its whole observation window lies inside that period. Public feeds list window-dated events only when the window is short; longer windows are shown as “changed between {date} and {date}”. The first live comparison for a state can span many months, for example when it is compared against an archived copy of the official file. Window-dated changes from that comparison appear in agency timelines in the product but are not counted as recent activity. Expiration dates are future dates and never count as past activity.

Event definitions

An event type is active in a state only when that state's source supports it. Inactive types are never shown as zero.

New agency: A license appeared for an organization not previously seen in any tracked source.
New license: A license record appeared in the official source (or, on a baseline run, has an issue date inside the lookback window).
New location: An existing organization gained a license at an address not previously associated with it.
Location closed: A license for one location became inactive or left the source while the organization keeps other active licenses.
License status change: The canonical license status changed, or the record was removed from the source.
Administrator change: The administrator named on the license changed. Only for sources that publish an administrator.
Licensee change: The licensee named on the license changed.
Address change: The normalized address on the license changed.

Product only: new license application, license renewed, license expiring, ownership change, phone change, name change. Pending license applications are never shown or counted on public pages.

Metric definitions

Agencies tracked: organizations with at least one active, relevant license, as of the last update.
New agencies (30 / 90 days): organizations with a New agency or New license event dated inside the trailing window, counted once each. Shown only once HomeCareIntent has observed the full window.
Licensing events: public event types dated inside the window under the dating rules above.
New agencies per month: by effective month. Months before tracking began are shown as “Not tracked”, never as zero.
Suppression: city-level counts below 3 are shown as “Fewer than 3”. State-level counts are exact.
Rankings: only cities with at least 3 organizations; ties listed alphabetically.
Percent change: only when the prior period is large enough to be meaningful and both periods were fully observed.
Multi-state figures: stated as “across N live states”, never as a U.S. total.

Coverage statuses and Live criteria

A jurisdiction becomes Live only after all of these pass and are recorded:

Official source identified and its regulator recorded.
Access reviewed: terms and robots rules respected, no login or CAPTCHA circumvention.
Schema documented.
Records normalize and pass validation.
Identifiers are reliable.
Ingestion is repeatable across consecutive scheduled runs.
Change detection is verified against the source by hand.
Freshness understood: source cadence recorded or marked “not stated”; check frequency set.
Quality checks pass on recent runs.
A sample of organizations is checked by hand against the source.

Live: Official source verified, ingestion repeatable, output manually reviewed. Figures are published.
Beta: Ingesting and under review. Figures are published with a Beta label and are never added to live totals.
In development: Connector in development and output under review. No figures are published.
Researching: The official source and access method are being evaluated. No figures are published.
Planned: On the national roadmap. Source research has not started or is incomplete.
Degraded: A quality check failed or the source changed. Publishing is paused until it is resolved.
Offline: The source is unavailable. No new data.

Quality controls

Every run is checked automatically for problems such as an empty or truncated source, unusual changes in record counts and changes to the file's structure. A failure halts publication for that run and can move the jurisdiction to Degraded. Records that disappear from a source are flagged for review rather than presented as closures unless the source confirms the closure.

Publication standards

State pages enter search indexes only when the state is Live, enough organizations are tracked to support the figures, the data is current and enough history exists. A page that falls short says why and stays out of search indexes.
City and new-agency pages follow the same principle: small cities are shown without indexing or not at all, so no page is published that the data cannot support.
Research reports publish only when every included Live state has sufficient history and was fully observed for the whole period. Reports are frozen at their publication date.

Intent score

The intent score is a deterministic, rule-based score from 0 to 100. It rises with recent, commercially relevant changes such as new licenses, new locations and leadership changes, more so when several occur close together, and reflects whether the organization can be reached. It is HomeCareIntent's interpretation, not a source fact, appears only in the product, and always lists the signals that contributed. No language model or hidden factor contributes.

Limitations

Terminology and license scope differ by state; cross-state comparisons use mapped categories.
Sources can lag reality; a change appears only after the source publishes it.
Not every state licenses non-medical home care, so some agencies appear in no state record.
Coverage is partial. Figures for Live states are not national totals.
Some sources publish no dates; those changes are window-dated as described above.

What we don't do

No patient, client or caregiver data. No protected health information.
No consumer reports and no decisions about individuals (credit, employment, insurance, housing).
No figures from sample data on public pages.
No language-model-written statistics or narratives; public sentences are computed templates.
No circumvention of logins, CAPTCHAs or robots rules.

Corrections

If a record looks wrong, tell us through the contact page with the page URL and what the official source shows. We review it against the stored snapshots. Original source history is never altered; a corrected event is retracted with a reason, and affected pages and reports carry a dated correction note.

Definitions

Home care (non-medical): Help with activities of daily living and household tasks, such as bathing, dressing, meals and companionship, that does not require a licensed clinician. States call these providers home care organizations, home care agencies, personal care agencies, personal assistance services agencies, companion or homemaker agencies, among other terms.
Home health agency: A provider of skilled clinical services at home, such as nursing or therapy, usually licensed separately and often Medicare-certified. Included only where a state licenses it under the same framework and the license type is marked relevant.
Home care organization (HCO): California's term for a licensed non-medical home care provider. Other states use other terms; each state page uses that state's own term.
Personal care agency / personal assistance services: Terms several states use for non-medical, hands-on personal care providers. Scope varies by state.
Medicare-certified home health agency: A home health agency certified to bill Medicare and listed in CMS public files. Federal data covers this subset only, not non-medical home care.
Organization: A single business entity resolved across records. It may hold several licenses and locations.
Location: A licensed site or branch of an organization.
License: A state-issued authorization record with a type, a status and, where published, dates.
Tracked organization: An organization with at least one active, relevant license in a Live or Beta jurisdiction as of the stated date.
New agency: An organization with a New agency or New license event dated inside the window: first detected after the baseline, or carrying an official issue date inside the window. Counted once per organization.
Signal / event: A change detected between two source snapshots, with its source, detected date and effective date.
Detected date: When HomeCareIntent first observed the change.
Effective date: When the source says the change took effect. When the source publishes no date, the event is window-dated (see How events are dated).
Source-dated event: An event whose date comes from the source, such as a license issue date.
Window-dated event: An event the source does not date. It happened at some point between the previous snapshot and the snapshot where HomeCareIntent saw it.
Baseline: The first complete ingestion of a source. It establishes the starting state and does not create new-agency events, except issue-date backfill.
Intent score: HomeCareIntent's deterministic 0–100 score combining recent events and contactability. An interpretation, not a source fact. Components are shown with every score in the product.
Source cadence vs check frequency: How often the government source publishes updates, versus how often HomeCareIntent checks it. They are always stated separately.

Changelog

2026-10-05 · Methodology published. Event dating rules: window-dated events count in a period only when their whole observation window lies inside it.
2026-10-05 · Wording condensed to describe each step at the level a reader needs. No change to how any figure is computed.