I welcome the discussion. My instinct is not always right, and the best way to find out if I'm right or wrong is to argue with someone about it. :-)
What's the difference between a new card and a card you've forgotten? If I'm overwhelmed by the workload, I don't want to introduce new cards or forgotten cards because they're both the same amount of work.
I think they're quite different workloads. A new card requires you to memorize new information. During the course of that process we may forget multiple times, but it's not always complete memory loss. A more realistic pattern is "have no idea" -> "remember vaguely" -> "half remembered" -> "almost remembered" -> "answered correctly". If you're at the "have no idea" stage then the card is effectively new, but all the other stages involve some degree of previous investment. It makes sense to keep being asked again, instead of going back to the start.
You say that it's just giving them a lower priority and not giving up on them, but delaying the already most-delayed cards is reducing whatever chance you may have of answering them to practically 0.
If a card has a long relative delay (say for example a card with a one day interval that was scheduled for the first day of your one week holiday), then I think that 2% would probably be more appropriate than 33%. I believe that cards like these should definitely be given a back seat.
The changes to the scheduler came about while chatting to a guy on #anki. He brought up the idea of a cutoff point, where cards are declared automatically failed (and thus not prioritized) if they have reached a certain level of delay.
I argued that such a cutoff point is hard to define, and you risk wasted effort if cards that you are just able to remember are deprioritized. I asked him to define a function which would produce an appropriate cutoff point, and after a little while he decided that perhaps it wasn't necessary for now.
Another thing to remember is that the "initial period" is not only for recently added cards. Failures further down the line (when you've remembered a card for 9 days, 30 days, even a year) can all result in the card returning to the initial state (automatically if you press '1' at an interval of less than about 30 days, and manually if you ever choose '0'). These are cards with a small interval, but where the likelyhood of recall past that interval is much greater. For example, after I implemented the scheduling changes, I started going through my deck. The first 5 cards were all at 3-5 day intervals, and had a delay of 30-40 days. The delay was about 8 times as much as the interval! And yet, 4/5 of the cards I was able to answer correctly. All of those cards were relatively new (never had an interval over 9), but I had failed them a number of times already and the memory had begun to cement itself.
If you go through the cards in a deck looking at the failed cards, I think you'll find that repetitions > 2 is common. Failing is not something which only happens before you "learn" a card - it's an inevitable part of the SRS process and regressions can happen to both mature and young cards.
There is one situation where I think that you would end up with a lot of failed, newly seen cards. That is when learning new material. But you can only learn new material when all existing material has been cleared, so the only situation where prioritizing failed cards may be inappropriate is where you review (and fail) 200 new cards, then go on holiday. In all likelyhood you'll never do this, as failed cards popping up every 10 minutes limits how far into a stack of new cards you can proceed before you're prevented from looking at any more. However, if you try and shoot through the deck as quickly as possible, failing everything, then perhaps it's possible. So, unlikely as it is, let's consider that you go on holiday after doing that.
In this case, a 2 week holiday may expire a number of other cards, and your concern is that those expired cards may be more important than the recently failed cards.
In the current version of Anki, this was never addressed directly - if a different card expired near the start of the deck, it would be shown before a failed card. If it expired after a failed card, it would be shown after it. Thus it still was not possible to finish all the expired cards first - you had to finish all reviews to be sure everything that needed to be reviewed was reviewed.
Anyway, this situation is a corner case, I think - and not something you're going to have to deal with normally. And if you did have to deal with it, why couldn't you just go into the editor, sort by added time, and suspend all the recently failed cards until after you got through the backlog?
Last edited by resolve (2007 October 22, 11:01 am)
