Experimentation
Sticky variant assignment
A visitor who saw variant B keeps seeing variant B. Anything else is a broken experience and a corrupted measurement.
Assignment happens once. The variant a visitor is given is recorded in their browser, keyed by experiment, and reused on every later page and every later visit from that browser.
Stickiness is a correctness requirement before it is a nicety. A visitor who is re-randomised on each pageview sees a page that changes under them, and contributes evidence to both arms at once, which makes the comparison meaningless.
The record is kept in browser storage rather than a cookie. That keeps the platform out of cookie consent while still giving each visitor a stable experience, and it is why the tracker sets nothing.
The store is pruned as experiments finish, so a long-lived visitor does not accumulate assignments for tests that ended a year ago. Assignments made by an older version of the runtime are migrated forward rather than dropped, so returning visitors keep the variant they were already seeing.