• have you every heard of Classical Test Theory
  • check ⧸r⧸anki, people there discussing about the 'traps' of not clicking buttons because the day nr there doesn't appeal and shit like that
    • → ~ don't expose your inner workings to the user
    • may be partly self-determination, users want to "hack the system"
      • which would then loop back into motivation in SR

• but for real, what are we measuring; ◊ recall, confidence, ⌣ cognitive load, recall speed, perceived difficulty, what more?

  • do these correlate?
  • how do you usually measure these? check papers

• Correct probably means 90% correctness often, not 100%. Is it then correct to bucket it in W=1?

  • And worse, Wrong definitely wildly differs between meaning 0%, 20%, 90%
    • say I have an two word italian sentence that I'm trying to memorize, and the true recall probability for each word is 50%, what do we want the algo to predict?
      • tbf this example is partly just "bad" non-◊ atomic learning content, but this kind of stuff even happens with single letters, syllables, grammatical gender, ...
      • also this is a kind of random model of memory, not sure if supportable

spaced-repetition