Skip to main content

Limit Feature Flag Exposure

Last updated on

When you create a feature flag, you can control how much traffic is evaluated by its targeting rules using the Limit exposure button in the Targeting rules section.

Limiting exposure means only a percentage of users are routed through the flag's targeting rules. The remaining users are excluded from evaluation and receive the default treatment, bypassing all targeting logic. This is useful for gradually ramping experiments or safely rolling out features.

  1. From the FME navigation menu, navigate to Feature Flags and click on a feature flag you want to configure.
  2. On the Definition tab, click Initiate Environment for an environment and go to the Targeting section.
  3. Click Limit exposure.
  4. Set the percentage of traffic that should be evaluated by the targeting rules.

Traffic allocation​

Users within the exposure percentage are evaluated against targeting rules and assigned a treatment. Users outside the exposure percentage are not evaluated against targeting rules and receive the default treatment. They are labeled in impressions as not in Split.

Increasing or decreasing the exposure percentage adjusts how many users enter an experiment, but does not change assignment for users already evaluated.

Sticky assignment​

Once a user is assigned a treatment within the exposed population, that assignment is sticky. Increasing exposure does not reshuffle users between treatments.

No re-bucketing on ramp​

Users are only added to or removed from the evaluated population. Treatment assignment only occurs within the exposed population and does not change when exposure is adjusted.

Ramp experiment traffic​

For multi-treatment experiments, keep treatment distribution stable while ramping exposure.

tip

Changing limit exposure or targeting rules creates a new feature version. For analysis, use the latest version once full ramp exposure is reached.

For example, with three treatments, you set the limit exposure to 20%. Treatment distribution is A (control) 33%, B 34%, and C 33%. This results in ~6.6% of total traffic per treatment, with 20% participating in the experiment and 80% receiving the default treatment.

This approach minimizes treatment crossover, prevents users from being reallocated between treatments during ramp-up, and preserves consistent statistical measurement over time. Unallocated traffic still triggers getTreatment, but is not part of any targeting rule and should not be used for metric comparisons.