Split testing
Split testing is Pro.
Arms and weights
A step can have variants. Each takes a percentage of the traffic; whatever the weights leave over goes to the step's own template, which is the control.
Two variants at 30 and 20 means half the traffic sees the original — deliberately, because a test that leaves no control is not a test. The control's share is shown while you edit, and if the variants between them ask for more than 100% the control share goes negative and the control panel says so rather than quietly normalising the numbers. Somebody who typed 60 and 60 meant something, and it was not 50/50.
A variant can name its own template, or leave it empty and inherit the step's — which is how a settings-only or copy-only test works.
Assignment is sticky
A visitor is drawn into an arm the first time they see the step, and the assignment is stored on their flow session. Reloading, coming back tomorrow, or moving forward and back again all give them the same arm. Assignment to the control is stored too, so it survives the same way.
If the arm they were given is later deleted or disabled, they are redrawn.
Reading the result
Fjord → Reports lists every arm of every split-tested step in the flow, with visitors, conversions, conversion rate, revenue per visitor, uplift over the control, and a two-proportion z-score.
Significance is reported at the usual two-tailed 95% threshold, and marked Too early until both arms have at least 30 visitors — below that the normal approximation does not mean anything, and a green tick on twelve visitors is worse than no tick at all.
Fjord does not pick winners. It will not reweight traffic, pause an arm, or promote a variant on its own. A plugin that silently changed what customers see on a number this thin would be harder to trust than one that reports it and leaves the call to a person.
Ending a test
Delete the losing variants, or set the winner's template on the step itself and remove the arms. Deleted variants keep their recorded events, so the report of what happened does not lose half the test the moment you act on it.