Ok, after dissecting the queries, it’s clear that the opposite approach is better, to seed an exposures table with all known user IDs. I’m using a single timestamp for all of them, one experiment name of course, and I tried one variation name, and ran the queries manually in my DW, so now I’m getting a main_sum and main_sum_squares. But when I rerun the exp results in GB, I still get ‘no data’ for all fields in my goal metrics regardless of statistics engine (probably b/c I’m only including one variation? I am ignoring conversion windows.) I know there are more queries or calcs happening behind the scenes, but I’m wondering if I do need to include two variation_ids in my seed exposures table after all, and how that will play out for this type of scenario given my variations are essentially time-based. Seems to borrow bits from feature flag and holdout experiments without fitting neatly into either. (And, FTR, I totally get it, this is not the standard use case, and we will undoubtedly have more standard use cases/exps that do follow more normal setups.)