treat the assignment logger as required, not optional, for anything you will ever analyze as an experiment. At SDK init, pass an assignmentLogger that takes the assignment object and writes it to your warehouse, for example a Segment track call with event name Eppo Randomized Assignment and the assignment properties. Verify end to end before launch: trigger an assignment, then query the warehouse for the event with the right experiment and variation fields. If you only add the logger later, the exposure period before that is unrecoverable, so wire it on day one even for flags you think are just toggles today.