Methodology

Each source in the registry is harvested in full (or by join key, for building records) through a generic connector. The connector validates the live schema against the fields the adapter expects, retries on rate limits, and stores every raw record with its retrieval time. Each run is diffed against the last, so additions, changes and removals are recorded; a record that leaves a feed is marked removed, never deleted.

Adapters map raw rows to canonical entities: facilities, operators, inspections, violation types and occurrences, buildings and tax lots. Every material fact carries provenance — source, source record ID, URL, retrieval date, effective date, jurisdiction and an evidence class. The build then generates geographic pages, reverse indexes and search shards, validates them, and publishes versioned JSON. The website reads only that published index.

Stable source identifiers come first: NYC DCID and permit number, OCFS facility ID, NYC BIN and BBL. We never merge records on name alone. NYC and NYS programs at the same normalized address are listed as possible same-site matches and kept separate. Operators are grouped only when published legal names match exactly after normalization; a parent company appears only with a cited source.

Texas publishes no coordinates, so we geocode center addresses with the US Census Geocoder and record the match quality. Home-based programs in Texas are shown by city and ZIP only, never pinned on the map. Home-based family day care providers are individuals. Their names stay on the regulator’s record and are not published as operator entities.

A violation type is a normalized jurisdiction code, such as NYC Health Code §47.43(f). An occurrence is one citation of that code at one facility on one inspection date, with its original code, severity, status and source row. Topic categories (supervision, staffing, safety…) are Inferred groupings over the regulation text. Codes from different jurisdictions are never treated as equivalent without a documented mapping.

NYC Department of Buildings violations, ECB violations and permits are linked to the building (BIN) that houses a program. They describe the property, are shown separately, and are never attributed to the childcare operator.

California CCLD publishes one statewide download of licensed child care centers with every licensing visit since 2021 (inspections, complaint investigations and other visits), each citation code with its plan-of-correction due date, Type A and Type B citation totals per visit kind, and each complaint’s substantiated, inconclusive and unfounded allegation counts. The download does not say which visit a citation came from, so we attach each citation to the latest visit on or before its correction due date and label it as matched. Coordinates come from the CDSS facility map layer; citation headings come from Title 22 of the California Code of Regulations. Family child care homes and the facility administrator’s name are not published here.

Texas HHSC publishes current inspections, report investigations and assessments since October 2017, with each cited standard, its risk level and the inspector’s narrative; these are updated daily.

NYC’s open inspection dataset stopped updating in April 2023. For current NYC center inspections we read each licensed center’s public page on Child Care Connect, the NYC Health Department’s own lookup: about the last three years of inspections, every violation with its official severity and OPEN or CORRECTED status, the permit expiration and the water lead test. We read it politely, one page at a time, and refresh it regularly. Inspections from the 2020–2023 open dataset are kept and labelled historical. Some violations may still be challenged at a hearing; the city notes that not all are final.

• VerifiedDirectly stated by the official source record.
• CorroboratedLinked by an exact identifier, with a minor discrepancy noted.
• InferredDerived by our method (e.g. topic grouping), not stated by the source.
• UnverifiedA possible link we have not confirmed. Not merged.
• HistoricalTrue as of a past date. Not current status.
• Not foundWe looked and found no record. This is not evidence that no issue exists.
• Not publicThe regulator does not publish this as open data.

Missing data never means no issue exists. A successful empty query, a failed query and an unavailable dataset are recorded separately. Proximity to an environmental site is not exposure. Consolidated parent-company financials are not facility revenue.