Delayed Experiment & Feature Gate Results for Warehouse-Native (WHN) Customers

Final update

-- Only WHN customers are impacted -- Around 2026-08-26 2:30 PT (16:30 UTC), we were experiencing a delay in processing experiment and feature gate exposure data for Warehouse-Native (WHN) customers. As a result, experiment and feature gate results may show fewer participants than expected for August 25–26, and cumulative metric totals from August 25 onward may appear understated. Live event ingestion and the Exposure Stream are unaffected. Data is delayed, not lost. As of 2026-08-27, 06:53 PT, our team has identified the root cause and actively worked to backfill affected data and restored normal processing. The underlying data has now been reprocessed and delivered. - If your experiments are on a scheduled reload, no action is needed , the next scheduled run recalculates the affected dates automatically and restores the missing participants. Results should be complete after that run. - A reload is required in two cases. If an experiment/gate has already been stopped or concluded, its results will not update on their own and need a full reload. If you reload manually rather than on a schedule, please run a reload before 28 August so the recalculation covers both affected dates;

1. Resolved · Aug 26, 09:30 UTC
   -- Only WHN customers are impacted -- Around 2026-08-26 2:30 PT (16:30 UTC), we were experiencing a delay in processing experiment and feature gate exposure data for Warehouse-Native (WHN) customers. As a result, experiment and feature gate results may show fewer participants than expected for August 25–26, and cumulative metric totals from August 25 onward may appear understated. Live event ingestion and the Exposure Stream are unaffected. Data is delayed, not lost. As of 2026-08-27, 06:53 PT, our team has identified the root cause and actively worked to backfill affected data and restored normal processing. The underlying data has now been reprocessed and delivered. - If your experiments are on a scheduled reload, no action is needed , the next scheduled run recalculates the affected dates automatically and restores the missing participants. Results should be complete after that run. - A reload is required in two cases. If an experiment/gate has already been stopped or concluded, its results will not update on their own and need a full reload. If you reload manually rather than on a schedule, please run a reload before 28 August so the recalculation covers both affected dates; after that a full reload is needed instead of an incremental one.