@seamus5,

OK... I've just spent a few hours mucking around with things on the backend to try and improve our default settings. This has resulted in a few changes that I think should improve the quality of the test and eliminate some of these frustrations. Specifically:

(1) previously we required people to answer words correctly 3 times before the words would enter delayed SRS testing. This could result in someone getting tested on the same word three times in succession before it would go away. To avoid this we've just reduced the number of forced initial tests to 2 times. So words will enter delayed testing faster now.

(2) the order in which we were pulling content from our database for testing was also not properly randomized, which meant that newly-added vocabulary was far more likely to be tested than older words. We want to have a slight bias towards newer materials since we can't rely on everyone to keep their SRS lists clean, but I've added a probabilistic check which will do a better job scrambling test options the majority of the time. So there is still a bias towards new materials, but not an overpowering one.

I'm also taking a look at tweaking our algorithms to enforce more variety in the initial testing circuit. Since we tend to default to English language tests, one idea is defaulting to a pinyin-test as our second test of word recognition assuming user settings permit. I do not want to add forced generative testing since it is much harder for people using mobile devices to quickly review content using those test options.