@SyHoekstra it was always a weird balance, when from basically day 2 of captchas we were using humans to train robots to read things that robots couldn't currently read. Given the humans read it incorrectly a lot of the time, it was never hard to get a robot to give the right answer in an acceptable number of tries.