Original research

Citation Lab editions

An experiment is a pre-registered before/after of one system-stamped ship against matched questions on the client's own board. Treatment moments are stamped by code, never by a human or a model. After the ship there is a 48-hour hypothesis window, then the design freezes. The estimand is a difference-in-differences with matched board controls. Verdicts are labelled attributable. They are not a claim that the treatment produced the move.

Every figure is measured on the named instrument versions; no client is named; the full methodology and dataset are on each edition.

No edition has been published yet. A month publishes only after enough opted-in sites have decided experiments to clear the floor.