Methodology
Where figures come from
Every statistic published on this site is the result of a SQL query that is checked into the repository next to the text that uses it. The text is written around the values the query returned. Nothing is written from memory, rounded by hand, or estimated.
Before a change is published, continuous integration re-runs each query and compares every number in the rendered page against the recorded result. A page whose text disagrees with its query does not ship.
Snapshots and provenance
Each ingestion writes an immutable, dated copy of exactly what the publisher served to object storage before anything is normalized. Those snapshots are never deleted, so any published figure can be traced back to the bytes it came from. Normalized tables are rebuilt from snapshots, never edited in place.
Freshness
A scheduled check compares the latest period each publisher offers against the latest period held here. When a source falls behind, changes shape, or stops responding, the check fails loudly rather than serving stale data quietly. Each page states the date of the snapshot behind it.
What we will not do
- We do not describe a correlation as a cause. Administrative data records what was reported, not why it happened.
- We do not compare jurisdictions without saying what the denominator is, or whether the set of reporting agencies changed between periods.
- We do not publish a page that has no data value of its own. Thin permutations are not generated, and where one exists it is excluded from search indexes.
- We reproduce a publisher’s disclaimer verbatim wherever the publisher requires it, including adverse-event reporting data.
- We do not use datasets whose licence forbids commercial use.
Corrections
Corrections are published with a date and a description of what changed, at /corrections, and are never removed.
By topic
Each topic carries its own sources and its own limits. Those limits are part of the method, not a caveat appended to it.
Federal spending
Federal awards, contracts, grants, and agency obligations, as reported by agencies to USAspending.
Sources
- USAspending API v2 — U.S. Department of the Treasury, Bureau of the Fiscal Service · U.S. Government work, public domain · updated daily, with monthly agency submissions
What this data cannot show
- USAspending reports obligations and outlays as submitted by agencies. It is not an audited financial statement.
- Award records are revised after publication; a figure is only as current as the snapshot it was drawn from.
- Place-of-performance is the recipient’s reported location, not necessarily where work occurred.
Crime and justice
Offense and arrest counts reported by law enforcement agencies to the FBI’s Crime Data Explorer.
Sources
- FBI Crime Data Explorer API — Federal Bureau of Investigation, Criminal Justice Information Services Division · U.S. Government work, public domain · updated annual release with quarterly updates
What this data cannot show
- Reporting to the FBI is voluntary. Coverage varies by agency and by year, so a change in a total can be a change in who reported.
- Counts are reported offenses, not convictions, and not a measure of crime that went unreported.
- Agency-level figures are not comparable across jurisdictions of different sizes without a population denominator.
Immigration and the border
Encounters, apprehensions, and enforcement actions published by U.S. Customs and Border Protection.
Sources
- Nationwide Encounters — U.S. Customs and Border Protection · U.S. Government work, public domain · updated monthly
What this data cannot show
- An encounter is an event, not a person. The same person may be encountered more than once in a period.
- CBP revises published monthly files; a figure is only as current as the snapshot it was drawn from.
- Encounter counts describe enforcement activity. They do not measure migration or the size of any population.
Health
Drug approvals, adverse-event reports, provider records, and hospital data published by federal health agencies.
Sources
- openFDA — U.S. Food and Drug Administration · U.S. Government work, public domain · updated weekly to quarterly by endpoint
- CMS Provider Data — Centers for Medicare & Medicaid Services · U.S. Government work, public domain · updated varies by dataset
What this data cannot show
- Adverse-event reports are voluntary, unverified submissions. A report does not establish that a product caused the event. Publisher disclaimers are reproduced verbatim on every page that uses this data.
- Report volume tracks reporting behaviour, publicity, and product uptake, not only product safety.
- Provider payment records describe billing, not the care that was delivered.
Nonprofits
Tax-exempt organizations as they appear in IRS Form 990 filings and the Exempt Organizations Business Master File.
Sources
- IRS Form 990 filings and Exempt Organizations Business Master File — Internal Revenue Service · U.S. Government work, public domain · updated monthly (bmf), rolling (filings)
What this data cannot show
- Form 990 data is self-reported by the filer and is not audited by the IRS before publication.
- Filings lag the fiscal year they describe, often by more than a year.
- The Business Master File records registration status, not whether an organization is active.
Page built