[yermierc]

We're starting with the assumption that learners will not be able to religiously follow a schedule. That means that whether you're cramming, checking in sporadically, or completing a daily queue - the algorithms should ideally adapt accordingly without making your "to do" list untenable. One component of this is determining whether your long-term retention is bounded to a specific timeframe (e.g. learn content X by future date Y) or unbounded (lifelong learning).

What we're borrowing is the concept of memory decay, but I don't think there are many similarities beyond that. Imagine learning spanish in college: you could set a goal for your final exam but also have a lifelong learning goal. Whether you're following your schedule or cramming, your answer events will communicate with the intersecting algorithms so that you're being optimized for several different use cases simultaneously. When we launch the "goal setting" feature, gjcourt may write a blog post about it and post it to HN...