Reproducibility and data provenance
The Atlas is a public-data exploration platform and research-support tool. It is not a completed epidemiological study, a causal analysis, a medical system, a diagnostic tool or a proof engine. Everything published here is designed so that another person can retrieve the same official source, apply the same documented transformation, and obtain the same view.
Application version atlas-7.0.0 · pipeline pipeline-2026.08
1. Provenance recorded for every published item
Every published map metric, environmental layer, comparison, chart, download and assistant answer carries the following record. Where the publisher does not supply a field, the Atlas says so rather than filling it in.
- Applies to
- The map metric, layer, chart or download the record describes.
- Dataset publisher
- The agency or organisation that released the data.
- Dataset title
- The publisher's own title for the dataset.
- Official source URL
- Link to the publisher's page for the data.
- Methods / data dictionary
- Link to the publisher's methods or data dictionary, where one exists.
- Source type
- Official government, state/local agency, peer-reviewed, university, nonprofit/community, journalism, or book/reference.
- Evidence tier
- How much weight the source may carry. Never a statement about causation.
- Source release date
- The date the publisher released this version.
- Download / import date
- The date the Atlas downloaded or imported it.
- Data time coverage
- The period the data describes.
- Geographic coverage
- The areas the release covers.
- Geographic unit
- The unit values are reported for (state, county GEOID, point location).
- Source version
- The publisher's version or extract label.
- Raw file / API response checksum
- SHA-256 of the exact file or API response that was ingested.
- Transformation / pipeline version
- The Atlas pipeline version that normalised the records.
- Data release ID
- The Atlas release identifier the published rows belong to.
- Metric definition version
- The version of the metric definition used to compute the figure.
- Known limitations
- Documented limitations, in the publisher's terms where possible.
- Data status
- Final, provisional, demo, staging, retired, unavailable, restricted, or suppressed.
2. Source types and evidence tiers
3. Data status vocabulary
The publisher has closed and finalised this period. Values are stable unless the publisher revises them.
The publisher has released this period before it is complete. Figures may change and are labelled separately from final data.
Synthetic demonstration records. Not real outcomes, and never citable.
Ingested and under review. Not shown on the public map.
Withdrawn or superseded. Kept only for the reproducibility record.
The publisher does not release this measure for this geography or period.
Access is restricted by the publisher; only the description is shown.
The publisher withheld the value to protect confidentiality. It is not zero and is never reconstructed.
4. Final and provisional data policy
Newer data can take time to appear because death and birth records are received, coded, investigated, reviewed, and finalized. Provisional figures may change. The Atlas labels final and provisional data separately.
Final data
- Final data may be used in verified county maps and benchmark comparisons once validation has passed.
- Final data always shows the word Final and the publisher's release date.
Provisional data
- Provisional data always carries a visibly distinct Provisional badge and its as-of date.
- Provisional data is never silently mixed with final data in the same figure.
- Provisional data is excluded from published county-level association comparisons by default.
- Provisional data may appear only in an optional 'Recent provisional context' view, and only where the publisher's coverage and completeness notes support it.
5. Reproduce this view
Every published map view has a Reproduce this view action in the Research quality panel. It exports a JSON record, or copies a plain-text version, containing: the view URL and filter state, the selected geography, the selected outcome and layers, the time period, the health-metric and environmental-indicator definitions, the data release IDs and source versions, the map mode, the benchmark definition when one is used, the excluded/suppressed/unavailable counts, the query timestamp, the application and pipeline versions, and the non-causation note.
It never contains individual records, private source data, suppressed values, secrets or raw health records. The export is checked against a forbidden-field list before it is written, and it is blocked if anything sensitive-looking is present.
Non-causation note: the Atlas shows aggregate public records in the same place and period. Co-occurrence, proximity or overlap in an area does not establish exposure, individual risk, or causation, and cannot explain why any individual pregnancy, birth or infant outcome occurred.
6. Temporal alignment of environmental records
A facility visible in today's regulatory file is not evidence that it existed during an earlier health period. Each environmental record stores its start and end dates when published, the source reporting period, the source publication date, the location precision, and what kind of record it is:
Facility presence record
The publisher records that a facility is listed at this location. It is not a measurement of anything released, and not a statement about any period other than the one published.
Regulatory / compliance record
The publisher records a permit, inspection or compliance event. It describes the regulatory file, not environmental conditions at any home.
Measured monitoring result
The publisher recorded an instrument measurement at this monitor for its own reporting period. It describes that monitor's location only.
Cleanup / contamination record
The publisher records known contamination or a cleanup programme. Listing does not describe present-day conditions off site.
Reported incident
A reported event as filed with the publisher. Reporting completeness varies by state and year.
When timing is unknown the Atlas states: “Timing of this environmental record cannot be aligned to the selected health period.” and the record is excluded by default from any time-aligned area comparison. When timing overlaps it states: “Record timing overlaps the selected period. Overlap and proximity do not establish exposure or causation.”