Strove Internals.md

How Strove Works

If you’re familiar with spaced reptition, you’ll know that there are a lot of approaches out there, and a lot of research. Anki, SuperMemo, Quizlet, Duolingo etc, all have their own proprietary algorithms implementing some variation on learning curves. When prototyping Strove, one of the things I discovered is that many of these algorithms, require a lot of unique information to be stored. For instance, SM2 stores an easinessFactor, dueDate, and interval for each card a user wants to learn. This is fine if you’re confident in an algorithm, but I wanted a more general-purpose solution that would allow me to fine-tune lesson generation over time.

Instead, the Prisma model looks something like this:

model Card {
  id        String  
  createdAt DateTime 
  front     String
  back      String
  answers   Answer[]
}

model Answer {
  id          String  
  createdAt   DateTime 
  correctness Float
  card        Card 
  cardId      String
}

This minimal solution, where we just store an answer each time a user attempts a card, actually contains everything we need to know! Now, instead of storing state, as is traditionally recommended, we can derive it from each Card's Answers.

to be continued