Should experiments always have unique tracking keys?
Say I run an experiment that has some targeting conditions. Should my second experiment, that tests the same feature but with different targeting conditions have a different key?
f
fresh-football-47124
09/20/2023, 7:59 PM
There are some cases where you may want to intentionally use the same tracking key
it will cause the data to be mixed on the database side
m
mysterious-nail-58688
09/20/2023, 8:00 PM
Yes, it will require filtering by our test dates, right?
f
fresh-football-47124
09/20/2023, 8:00 PM
you can isolate that first case with phases to isolate by date, or just use the same key - or you can also use the same key, but re-randomize the existing users… it’s pretty flexible
if you want to throw away that initial data, changing the tracking key will keep things isolated and re-randomized
m
mysterious-nail-58688
09/20/2023, 8:08 PM
Does the re-randomization happen automatically with a new phase?
f
fresh-football-47124
09/20/2023, 8:35 PM
no
phases, are done on the reporting side, and are date partitions
but most actions which cause re-randomizations should cause a new phase
m
mysterious-nail-58688
09/20/2023, 8:37 PM
The randomization you are talking about is how we assign users to a test, right?