A citation tracer for statistics that travel. Each number is followed from the version everyone repeats back to the primary document — and to the exact point where the retelling drifts.
Exhibit A · how a statistic loses its footnote
Cook et al. (2013) read 11,944 abstracts. Two-thirds stated no position; of the third that did, 97.1% endorsed human-caused warming. The figure is real — and other studies put the consensus higher still. What dropped on the way to the headline was the phrase “expressing a position.” Read the primary study ↗
Open the register ↓The number in the form it actually circulates — speeches, posts, headlines, slides.
Follow it back to the study, agency, or filing it rests on — or find that it rests on nothing.
The primary text, verbatim and attributed. The words the popular version was built from.
The exact move — a dropped caveat, a widened scope, a swapped word — and what it changes.
What this is not. Provenance does not rule on whether an underlying issue is real or important. Climate consensus is real; hunger is real; gun deaths are real. The register checks one narrower thing: whether the cited number says what the citation implies it says. Often the honest correction makes a claim stronger, not weaker.
Eight ways a statistic drifts from its source. The codes below are both a taxonomy and a filter — most distortions are one of these moves, and a few are several at once.
Filter by drift type
Hover a code to see how that distortion works.
Filter by domain