Hi Growthbook team, There was an experiment which...
# ask-questions
w
Hi Growthbook team, There was an experiment which we concluded on 19th June. i've added gb dashboard image where last revision was done on -> Jun 19, 2026. where default value was updated and experiment related to it was disabled. but we are able to see exposures getting sent for this experiment. the number is pretty small but still we're not able to figure out how come for a feature flag with default scaled value and a disabled experiment is sending exposures. let us know if anything is needed from us to help debug this issue.
f
Hi @witty-rocket-75751 good morning If the experiment was disabled on June 19 and the feature default was updated, we wouldn’t expect new exposures for that experiment once clients have the latest feature payload. When we see a small number of exposures after that, the most common causes are stale/cached SDK payloads, older app/browser sessions still using a previous payload, delayed event ingestion, or events that were generated before the change but arrived later. Could you check a few things on your side? • Compare the exposure timestamp vs the received/ingestion timestamp for the events still coming through. • Confirm whether those exposures are from web, mobile, backend, or a specific SDK. • Check whether any clients may still be using a cached/older feature payload from before June 19. • If possible, inspect one affected user/session and confirm whether the SDK payload still contains the old experiment rule. • Confirm whether the exposures are coming from production only, or another environment as well. That should help confirm whether these are truly new evaluations after the experiment was disabled, or delayed/stale events from clients still running the older payload.
w
• the timestamps was for after 19th june only. • all are via mobile flutter SDK. • this shouldn't be the case. • how can we check this part? • all are production exposures.
f
Is the experiment only reachable as a rule in a feature flag, or is there also a non-flagged visual editor or URL redirect implementation in the experiment?
w
it was only reachable as a rule but as mentioned it was disbaled so ideally it should have been unreachable altogather.