⌬ birds-exposure-training

• https://github.com/koljapluemer/birds-exposure-training

Lessons Learned

• regarding auto-gen wrong answers:

  • trial also tells you something about learners knowledge of the other item
  • if you use ebisu, and a wrong item keeps coming up, pure frequency can give you a hint that this is the right answer
    • therefor, mix in the least-likely-remembered item as the wrong answer for a random new item

• main learnings:

  • binary choice in learning apps has implications for both items
  • if you repeat a learning item in multiple choice rapidly, user can correctly guess answer just by how often they see it
  • ~ use urgent or due items for all choices of multiple choice
  • ~ you need learning data to know what hard choices for multiple choices could be
  • tracking stuff like 'trailing success rate' or 'streak correct' doesn't make sense if SR is counteracting