Start with the source.
Every published observation should identify the source agency, product, release, table or field, reference period and transformation method. A link to an agency homepage alone does not explain where a particular number came from.
We separate source-native evidence from normalized values and display formatting. That preserves an audit trail when an agency revises a release or a parser changes.
Geography is part of the definition.
A city, county, metropolitan area and postal ZIP code describe different places. We preserve the source geography and its boundary vintage rather than relabeling a convenient number as city data.
Names help people find a place. Stable geographic identifiers distinguish places with the same name and keep a renamed place connected to its history. Census ZCTAs and postal ZIP codes are treated as separate concepts.
Show when a measure applies.
The observation period, agency release date, retrieval date and publication date answer different questions. Republish an unchanged statistic and its reference period stays the same.
A multi-year survey estimate is not a single-year count. Inflation-adjusted currency values need their dollar year. A trend should make these distinctions visible.
Unavailable never means zero.
A value can be estimated, suppressed, unavailable, not applicable, provisional or revised. These states need their own labels. We do not reconstruct suppressed cells through a derived calculation.
Sampling uncertainty and coverage limitations belong beside the answer. Where a supported significance calculation is unavailable, a difference can be described without claiming it is statistically meaningful.
Check the inputs before the difference.
Comparisons must consider the metric definition, denominator, units, native geography, boundary vintage, observation window, adjustment and source methodology. Incompatible inputs should produce an explanation rather than a misleading percentage.
We do not average medians or sum incompatible rates. A ranking also needs a defined eligible universe, inclusion rules and a reproducible method.
Review once, publish consistently.
The intended publication unit is a reviewed snapshot. Profiles, comparison tools, exports and APIs should read the same approved observations. Source rights, attribution and redistribution permissions are checked before the relevant distribution is enabled.
The current site is an early implementation. Development profiles identify their source and review status. Some may show source-linked observations for verification; this does not imply approval as a public pilot. This page describes the standards the product is being built to enforce.
See our source approach