Experiments

import ScaleBun from '@scalebun/react-native';
ScaleBun.experiments.<method>(…);

Members#

Reference#

variant#

ScaleBun.experiments.variant(key: string): string | null

Notes#

experiments.variant(key) returns the variant this user is assigned to, or null when they are not in the experiment. It is the client half of the A/B and multivariate tests configured in the dashboard.

null is a real answer and needs handling: it means no assignment — the experiment is not running, the key is unknown, or configuration has not arrived yet. Treat it as "show the control".

Assignment must be read, not computed#

Do not derive a variant yourself from a user id or a random number. Assignment comes from the delivered configuration so that the dashboard and your app agree about who saw what. A locally computed variant makes experiment analytics meaningless, because the reported arms will not match the rendered ones.

Ordering#

1. 1

   init(), then identify(), then variant()

   Assignment stability depends on a stable identifier, the same as rollouts. An unidentified user is bucketed per device.

2. 2

   Read once per exposure, at the moment of exposure

   Not at launch, and not repeatedly in a render loop — hold the value in state for that screen.

3. 3

   Handle null as control

   Every call site needs that branch. It is the state before configuration arrives, which is exactly when your first screen renders.

Limits and edge cases#

• A user must stay in one arm. If assignment changes mid-experiment the results are invalid. This is why identity comes first.

• Propagation delay applies — starting an experiment reaches devices with their next configuration fetch.

• The metric must be tracked separately. This namespace assigns; it does not measure. Track the outcome event yourself.

• Client-side assignment is visible to anyone inspecting the app, including which variants exist.

Related#

• A/B tests

• Experiment analytics