Trust, explained
How THE FILE works
We turn validated observations from official public sources into readable profiles and reports. The evidence boundary comes before the headline.
01
Official sources
Real observations come directly from source agencies such as the U.S. Census Bureau, BEA, EIA, and BLS. Data.gov is used for catalog discovery, not as a substitute for the source observation.
02
Deterministic computation
Metrics, rankings, distributions, and comparisons are calculated by code before any story text is written. LLM output cannot create a number or change its meaning.
03
Fact packs and verification
A fact pack preserves the metric definition, geography, period, source lineage, caveats, and claim boundaries. Only published content with a passed machine verifier is shown publicly.
04
ACS estimates need care
Census ACS five-year values are estimates, not exact counts. A single point estimate does not establish a trend, cause, or statistically significant difference. Our current profile cards do not include margins of error, so we do not present significance claims.
05
What we refuse to infer
We do not turn separate measures into an overall state score, average state values to invent a national baseline, or describe a nominal ranking as proof that one place is meaningfully better.
06
Publication and corrections
Evidence and audit history remain append-only. Published content can be corrected, unpublished, or archived through the controlled publication path without rewriting the underlying evidence.