Missing or mistimed exposure: confirm the problem

An experiment result is trustworthy only if the exposure event fires when a user encounters the variant. Missing or early exposure events, or events from users who didn't see the variant, cause the analysis to measure the wrong population.

Data Assistant doesn't perform this experiment-specific check. Use Experiment diagnostics and individual user timelines to confirm exposure timing.

Use Experiment diagnostics for Amplitude experiments. If another system assigns variants, compare that system's assignment records with the exposure event in Amplitude.

Review aggregate diagnostics

1. Open the running experiment and find the Diagnostics card.
2. Review cumulative exposures for unexpected gaps, spikes, or diverging variant lines.
3. Review the assignment-to-exposure funnel when it appears.
4. Review Exposures without Assignments when it appears. The chart doesn't appear when the assignment event serves as the exposure event or the experiment uses local evaluation.
5. Segment [Experiment] Exposure by flag key in an Event Segmentation chart when you need to isolate one flag's trend.

Inspect representative users

1. Choose test users from each variant, including important identity transitions such as sign-in or account switching.
2. Go to Users > User Profiles and open each account's event timeline.
3. Confirm that assignment precedes exposure when your evaluation mode produces both events.
4. Confirm exposure fires when the user encounters the variant, not when code fetches an unused variant or returns a fallback value.
5. Compare the user_id and device_id values on assignment, exposure, and surrounding metric events.

Some assignment and exposure events arrive on different days. A delay is a problem if identity or targeting properties change between the events.

Exposure correction limits in Amplitude

Amplitude can't create or retime exposures that your implementation never sent. Changing an analysis setting doesn't recover missing or mistimed exposures. Use diagnostics and timelines to identify where exposure timing breaks so engineers can correct exposure for the applicable evaluation mode.

Send findings to the experiment owner

Correcting exposure requires a code or configuration change. Send engineers the findings. Copy this template and replace the brackets:

Our experiment [NAME / FLAG KEY] has an exposure problem. In diagnostics/user
timelines, [describe: exposure is missing / fires before the variant renders /
fires for users with no assignment / identity changes between assignment and
exposure]. Evaluation mode is [client / remote / server-side local]. Make
exposure fire only when the user actually encounters the variant, with the same
user_id and device_id as the metric events. Our experiment metrics are unreliable
until this is fixed.

If you have no engineering resources

Exposure tracking is instrumentation, so someone who can change the experiment integration must complete this work. Contact these resources in order:

• Your Amplitude account team or Customer Success Manager.
• A partner or agency that manages your experimentation setup.
• Amplitude Support at gethelp.amplitude.com for product questions.

To correct exposure for your evaluation mode, continue with the engineering solution.