For some experiments, assignment and exposure are separate events (e.g. you may need to generate a web page that has an experiment at the bottom of the page. Users are exposed only if the user scrolls down and sees the experiment).
If you are using an assignment tool that logs an exposure event before users are actually exposed to the experiment, you need to filter down the list of users in the assignment source to people who actually saw the experiment.
You can filter the assignment source data based on a Qualifying Event. You can filter assignments to only include subjects that have either generated (or not generated) this qualifying event.
You can find these settings in the Advanced options of the Setup page when creating an Analyze-only Experiment.
When using Statsig SDKs you can use getExperimentWithExposureLoggingDisabled
on assignment and manuallyLogExperimentExposure
at the point of exposure to accurately capture exposure. You don't need to "filter" or clean these exposures.