Seems like this one is a better candidate:
402 rows in set (0.07 sec)
Compare with the naive
402 rows in set (1.90 sec)
The current one does a join with t_optimize, but the intention is to get all active pairs, so I think we can skip that join.