In many cases with learning cards, the actual interval at which they’re shown is often longer than the set interval (e.g., being seen again in 5m when the Again interval was set to 30s). This is because of other learning cards ahead in the queue, which could be piling up after failing a lot of learning cards in a row or carrying over from the previous day. This can mean juggling up to 20+ cards at once to see a failed card that was supposed to be seen in 30 seconds; for example, it may now only be seen again in 5 minutes, once the 20+ card cycle is completed and the cycle starts again. This is incredibly draining.
If FSRS 7, which attempts to model short-term memory, were to come to fruition, it’s important that cards are actually shown at the intervals that are set (i.e., that 30s really means 30s).
Of course, integrating this is hard. No method would be perfect; it would all be attempts to minimize the discrepancy.
Ideas discussed thus far:
- Showing learning card X after Y number of cards have passed.
- Automatic resorting of the learning queue by retrievability, provided that R is updated every time a learning card is rated.
- Showing cards with more attempts first.
If people have other ideas, they’re welcome to share.