What this build is measuring
identity_override— how often someone rejects the identity the quiz gave them. High means the taxonomy is wrong.soldout_recovery— of people who open a sold piece, how many click an alternative instead of leaving. The core metric.borrow_start/borrow_end— which identities get borrowed, and from where. Tells you whether occasion mode is real.quiz_complete— target 65% of starts. Below that, fix structure before content.rail_depth— items live per identity per size, logged on every shop view. Below roughly 15 the rail feels empty.