Learning a language with Anki is effective and frustrating for precisely the same reason: they test vocabulary at just the right time, when you're just about to forget the word.
Review sessions become one long beating-yourself-up. "I know I know this!"
Conversation
This is a significant (and underappreciated) design challenge!
For me a key insight is that we haven’t found good representations of progress. You need to see/feel your overall progress; emphasizing individual items produces the problem you describe.
4
1
25
What's wrong with making a simple adjustment to test slightly earlier? It seems like somewhat reduced effectiveness is a fair price to pay for somewhat improved user satisfaction. Especially if that user satisfaction reduces drop-out, and thus improves global effectiveness!
1
Yes, that might be a reasonable strategy!
My analyses of Quantum Country suggest questions follow a strong power-law distribution; 12% of questions cause most lapses. Most reviews are probably already “wasted time”! Improving this would exacerbate the problem we’re discussing.
This is probably not (as) true of language learning. I’ve seen a few papers show distributions for e.g. vocab words, and they’re much closer to normal.