A publishing platform that cannot make up a number.

From raw data to a published claim

Four stages, the same for every article on the site.

Peer review, built into the machine

Before anything is computed at scale from a new source, we write down the full analysis plan — which estimates get precomputed, under which survey weights, with which demographic cuts, and what gets suppressed — and render it into a single readable document. A human statistician reviews the statistics in that plan: the survey-design handling, the cuts, the cell-size rules, the exact wording of what we'd claim.

The review is binding, not a courtesy read. The approval is tied to a fingerprint of the exact plan a reviewer read — edit the plan afterward, and the fingerprint breaks, reopening review. Small cells are pooled and the pooled range is labeled; we never lower a suppression floor to make a cut work. Series breaks — a survey redesign, a mode change, a weight splice — are shown as breaks, never drawn through as one continuous line.

One design mistake would replicate across thousands of published numbers. So a person signs every plan before a single number is computed.

There is deliberately a human at the other end of every one of these.