Data and Methodology

DaycareFacts.com is a public-record research tool. Our goal is to make official childcare inspection records easier to find, compare, and understand without turning those records into a score, endorsement, accusation, or legal conclusion.

The core rule is simple: we work from official source records, preserve links back to those records, and do not make up missing information. The goal is to stay as fact-based and transparent as possible.

Journalists, academic researchers, and independent writers can also use the DaycareFacts Data Desk to request a custom, source-linked analysis for a story or research project.

Our work combines public-record collection, structured data analysis, software classification, and parent-oriented interpretation. For questions that depend on formal regulatory meaning, licensing enforcement, or legal interpretation, official agency records and qualified regulatory guidance should control. We welcome feedback from people with childcare licensing or regulatory experience in any state, and in particular FL, CA or TX.

Data sources

Updates and date context

Records are periodically updated. We do not have a fixed update scheduled yet. Once we do, we will announce it here.

Because public records can change, provider pages show date, including when records were checked and the dates of the inspection reports being summarized. A provider's current conditions may have changed after the latest available inspection report, so report dates matter.

When reading any provider page, pay attention to three separate dates:

• the date DaycareFacts.com last checked or imported the available records
• the date of the latest inspection report
• the dates of any specific recorded violations shown on the page

Those dates help separate an older historical violation from a more recent one.

Data corrections

August 12, 2026 - Florida and California

DaycareFacts corrected deterministic processing of saved official records in both states. These were corrections to DaycareFacts' derived classifications and display counts, not changes to the underlying agency reports.

• Florida: CARES reports were reprocessed with report-text parser v6 and topic classifier v18. The correction improved topic and concern assignments when authoritative checklist identity or official finding text conflicted with secondary wording. The 11,810-provider, 95,532-report, and 67,724-finding totals were unchanged, and all 59,050 parent-question cards and their question prompts were preserved. One explicitly reviewed explanatory note was corrected to match its source finding. We recalculated What Florida Childcare Inspectors Cite Most Often - Published August 8, 2026 · Updated August 12, 2026. The separate finding-rate study did not change because its report and finding counts were unaffected.
• California: CCLD findings were reprocessed with evidence-aware classifier v21 so operative deficiency evidence is distinguished from regulation boilerplate and plan-of-correction text. This corrected overly broad topic matches, including some records incorrectly labeled sleep safety. Import adapter v3 also corrected repeated-topic violation counts that had displayed as zero. The 28,624-provider, 127,494-report, and 59,355-finding totals and official source text were unchanged.

What records are included

The site focuses on official inspection records available through the public source systems for childcare providers in the current DaycareFacts.com dataset.

The system includes inspection reports that can be connected back to a source provider and an official report document or official inspection record. It does not use private tips, parent reviews, star ratings, social media comments, marketing claims, or unofficial allegations as inspection findings.

The current public product focuses on provider listings, inspection history, recorded violations from official reports, violation topics and concern levels, repeated topics across multiple inspections, and links to official report documents.

Florida provider pages include "Questions to ask" as prompts for parent follow-up. Some were generated offline using Google Gemini from structured inspection findings and source-linked context, then reviewed and imported before publication. California and Texas provider pages do not currently show generated questions. Gemini is not called when provider pages load.

What a recorded violation means

"Recorded violation" is the parent-facing umbrella term DaycareFacts.com uses for an issue recorded in an official inspection report. Depending on the source, the agency may call the item a noncompliance, deficiency, citation, checklist finding, or violation.

The phrase is not an independent legal determination by DaycareFacts.com. It describes what an official report recorded and keeps that source attribution visible. In structured data and Research, the site also uses "finding" as a neutral cross-state term for one such source-record item.

A recorded violation does not automatically mean a provider is unsafe or establish present conditions or whether correction occurred. It means the official report recorded an issue that may be worth understanding in context. The useful questions are usually:

• What did the report say?
• How serious is the topic?
• How recent was it?
• Was it corrected at inspection or assigned a due date?
• Did the same topic appear in more than one inspection?
• Was there a later inspection with no recorded violations?

Key definitions

Recent means within the last 12 months, based on the records-checked date used for the provider metrics. Older recorded violations remain part of the inspection history, but they should be read with their dates in mind.

Repeated topic means the same violation topic appears in two or more separate inspections within a 36-month window. The repeated-topic rule looks for patterns across inspections, not just repeated wording inside one report. A topic appearing across separate inspections may be more useful for parents to notice and ask about.

Higher concern is a parent-facing classification used for topics that may have a more direct connection to child supervision, access, health, safety, or accountability. It is not a legal ranking and it is not an official regulatory category unless the source report itself uses that language.

Examples of higher-concern topics include staff-to-child ratio, supervision, background screening, medication, transportation, sleep safety, hazardous materials or unsafe access, pool or water-area barriers, and child guidance, discipline, abuse, or neglect-related report language.

Medium-concern topics generally include issues such as sanitation, food handling, emergency readiness, playground or equipment condition, staff training, health records, or incident documentation.

Low-concern topics generally include administrative posting, paperwork, and recordkeeping issues that do not show a direct child-safety concern in the available report language.

How classification works

The system maps recorded violations into a topic and concern level using deterministic software rules. "Deterministic" means the same input record should produce the same output label under the same algorithm version.

Florida rules prioritize authoritative checklist identity and official finding text for reviewed rule families before using secondary comments. California rules separate regulation text, operative deficiency evidence, and plan-of-correction text so that incidental words in boilerplate or correction instructions do not override the reported facts. Source-derived California Type A and Type B severity remains preserved.

Current public topic examples include:

• Staff-to-child ratio
• Supervision
• Background screening
• Transportation
• Medication
• Sleep safety
• Hazardous access
• Pool barrier
• Child guidance
• Health or food records
• Staff training
• Emergency preparedness
• Incident records
• Facility condition
• Recordkeeping
• Administrative posting

For public display, concern levels are shown as Higher concern, Medium concern, and Low concern.

California Type A and Type B classifications are preserved as official source details. DaycareFacts.com separately applies its deterministic, parent-facing concern topics and levels; those classifications are not scores or official regulatory ratings.

The site also uses compact provider-level labels to help parents scan results:

• No recorded violations
• Questions worth asking
• Recent higher-concern violation

Those labels are summaries, not scores. "No recorded violations" means the available reports did not produce displayed recorded violations for that provider. It does not mean the provider is certified safe or that current conditions have been independently verified. "Questions worth asking" means the records include topics parents may want to understand before enrolling. "Recent higher-concern violation" means at least one recorded violation classified at the Higher concern level appears in the recent 12-month window.

Corrections and later inspections

Correction status adds context, but it does not erase a recorded violation.

If a report says a violation was corrected at the time of inspection, the provider page shows that context when available. If a report gives a due date or other correction instruction, the page shows that when available. If a later inspection has no recorded violations, that later inspection is also useful context.

However, the original recorded violation remains part of the official inspection history. DaycareFacts.com does not delete or hide a recorded violation simply because it was corrected, and it does not imply that a later inspection with no recorded violations proves current safety. The goal is to present the sequence of records so parents can understand what happened over time.

Missing reports and parser limits

The system is designed not to invent data.

If an official source lists an inspection, visit, or action but the associated public document cannot be retrieved, the system treats that as a source limitation. It does not guess what the record or document said.

If a source document exists but the applicable parser cannot reliably extract a finding, topic, checklist section, or deficiency block, the normal quality-control process is to stop and review the parser or classifier rather than silently publish uncertain structured data. Narrow source-side omissions may be accepted when they are documented and limited, but parser or classifier failures are not treated as acceptable simply because they are inconvenient.

In plain English: if the source record is missing, we say it is missing. If our software cannot read something reliably, we do not pretend that it did.

Quality control

The data process uses several safeguards before records are imported and displayed.

First, the collection process preserves source records and report documents where available. This makes it possible to trace a public finding back to the official record that produced it.

Second, source-specific parsers extract structured findings from saved PDF text, HTML, text, and structured records. The data pipeline retains source references such as provider license, inspection ID, report date, source file path, source hash, and parsed block hash.

Third, the classifier assigns topics and concern levels through inspectable rules rather than free-form judgment. The rules can be reviewed, tested, and versioned.

Fourth, source batches and county coverage use quality gates. The process checks for unavailable source documents, missing required sections or deficiency blocks, parser errors, source hashes, classification records that require manual review, and California family-home privacy protections. Batches with systemic parser, classifier, or privacy issues are paused for correction before public use.

Finally, route and page checks verify that public pages load, provider pages link to source documents, Florida question cards render from reviewed imported data, and California provider pages do not expose generated questions or private location data.

AI boundaries

The public provider labels, counts, sorting, inspection dates, repeated-topic detection, and concern-level classifications are produced by deterministic software rules.

Florida provider pages include Questions to ask; some were generated offline with Google Gemini from structured inspection findings and source-linked context, then reviewed and imported before publication. California and Texas provider pages do not currently show generated questions. The live website does not call Gemini or any other large language model when provider pages load.

AI is not used to decide whether a report exists, count findings, sort providers, determine provider-level labels, or fill missing data. The site should never use AI to invent source facts.

Current methodology version

The current public methodology uses source-specific deterministic parsing, classification, and provider-metric rules. Texas classification reads the official standard heading and, where the state publishes one, the deficiency narrative.

Current internal versions as of August 22, 2026:

• Florida CARES report-text parser: cares_text_violation_parser_v6
• Florida CARES topic and concern classifier: cares_topic_severity_v18
• Florida CARES provider metrics: cares_provider_metrics_v1
• Florida CARES county batch process: cares_county_batch_v2
• California CCLD report parser: ca_ccld_report_html_offline_parser_v20
• California CCLD evidence classifier: ca_ccld_evidence_classifier_v21
• California CCLD provider metrics: ca_ccld_provider_metrics_offline_v1
• California CCLD import adapter: ca_ccld_import_adapter_dry_run_v3
• Texas HHSC topic and concern classifier: tx_hhsc_four_type_standard_heading_classifier_v1
• Texas HHSC classifier rule tables: tx_hhsc_standard_narrative_classifier_v1
• Texas HHSC parent-concern policy: tx_parent_concern_mapping_approved_20260817_v1
• Texas HHSC import projection: tx_hhsc_four_type_application_projection_v1

These version names are internal labels. They are useful because they make the methodology auditable. If the parser or taxonomy changes, older outputs can be distinguished from newer outputs.

Known limitations

Public inspection data is valuable, but it is not perfect.

Known limitations include:

• official source records may be incomplete, delayed, corrected, or temporarily unavailable;
• a provider's current conditions may differ from the latest available inspection report;
• inspection reports may not include every detail a parent would want to know;
• source record wording and format can vary by source, jurisdiction, or licensing authority;
• PDF, HTML, text, and structured-record extraction can be imperfect;
• software classification can make mistakes;
• the site does not make legal conclusions or official regulatory determinations;
• the site does not replace calling the provider, visiting the provider or childcare setting, or checking current licensing status with the appropriate agency.

For these reasons, DaycareFacts.com should be used as a research aid. It helps parents find and understand public records more efficiently, but it should not be the only step in a childcare decision.

Contact and corrections

If a provider, parent, regulator, or other reader believes a record is missing, stale, incorrectly classified, or incorrectly displayed, they can contact DaycareFacts.com for review.

Helpful correction reports include:

• provider name;
• license or provider number if known;
• county or city;
• inspection report date;
• the official source report link if available;
• a short description of what appears incorrect.

Correction requests are reviewed against official source records. If the public source record changes, the site can update its displayed data during a future periodic refresh or targeted correction review.

You can submit corrections or questions through the contact page.