Here is my 1,314 word essay on the subject. I would have cited my sources more than twice, but really, it’s mostly an exercise for my benefit.
Explaining SRS and it's core concepts.
With spaced repetition software (SRS), which WaniKani is a form of, we take advantage of a few things, but the main concepts here are: active recall, the spacing effect and desirable difficulty.
Active recall means that our memories are more efficiently consolidated in long-term memory by us recalling the information, as opposed to passive review which involves us being reminded of the information by reading or being told. [https://www.ncbi.nlm.nih.gov/pubmed/20804289]
The spacing effect shows that learning is more effective when study sessions are spaced out, rather than massed study (called “cramming”, also what most schools do in studying a new topic every day and never revisiting anything). Longer intervals have been shown to be better than short ones, probably due to increased difficulty.
Desirable difficulty shows that the more difficult it is a learning task is, the more effort required, the more beneficial it is to consolidating memory. Easy tasks could be said to give the illusion of learning, especially where it has been proven that the difficult way is better (re-reading notes is easier than using flash cards/SRS, and has been shown to be less effective), and so are a lesser use of time. Obviously, to be desirable it has to be complete-able. Therefore it makes sense to increase difficulty over time, rather than making 1 impossible task.
Combining these 3 things, we get the spaced repetition system (SRS), where we actively recall information over progressively longer intervals, thereby increasing the difficulty over time, consolidating the memory as much as possible. This is about as efficient as a fact-learning system as we can make with current research and technology.
There are other ways to improve learning, but they may be much more difficult to incorporate, and their benefits to memory consolidation may be tenuous, especially when we’re already improving efficiency with the above effects. This system is already pretty good, and still miles ahead of the typical educational system of “this is information, write it down 50 times, then we’ll review at the end of the year and assign your Human Worth Certifications”.
Forgetting an item, but advancing it anyway: bad.
If we cannot remember an item at a specific level, it is because the memory has not been consolidated to that level. Perhaps the interval was too long a jump (a 30 day interval to a 120 day interval is a big jump, 400% the previous interval). In any case, the item needs to be re-consolidated, by starting it again. (WaniKani only puts you back 2 levels, which isn’t very good. It makes it feel less punishing, but it’s actually cheating you out of rebuilding the memory and only saves 10% the time).
If we instead “redo” the item, putting in the answer we’ve just been told, we’re no longer using active recall, we’re being passively reminded and our memory will not be strengthened as well as it could. Not only that, but the difficulty was dropped to 0 by having the answer passively reviewed, weakening the improvement in memory we could have had further.
Example: let’s say we get Guru-2 correct. We have a memory-strength of 7 days. We wait 14 days. Then we get Master incorrect, but advance it anyway. The system now has our memory strength for that item down as Master, and will expect Master memory-strength (30 days, which we don’t have. We have a memory-strength of less than 14 days. Also, without the active recall and difficulty, our memory has had less strengthening).
So the link between the SRS and our actual memory has been broken. We’ll need to be very lucky to remember it next time (the evidence definitely suggests we won’t remember it next time), and if we continue to progress an item beyond it’s actual strength in our mind we’ll end up burning the item and we’ll never actually get it deep into long term memory.
Even if we don’t burn an item we obviously don’t know: allowing the item to continue is a waste of time. Because by restarting the memorisation process now, you get it out of the way, it’ll be done sooner. By delaying it until the next interval, you’ve added that interval’s time to the time taken to relearn the item.
Getting an item wrong, but advancing it anyway: very bad.
Basically the same as forgetting it, but much more insidious. Because we did use active recall, and the difficulty was there, the memory is strengthened a lot… except our answer was incorrect. We’re consolidating misinformation into our long-term memory.
It is therefore of dire importance that we re-consolidate the correct memory by starting it again. The incorrect memory needs to be overpowered by the correct memory, and the only way to be certain of that is to give the correct memory all the consolidation we can (by starting again).
Getting an item correct, but inputting it wrong, still advancing it: good.
Valid times to re-enter an answer: typos and knowing what you meant to put in. Because your memory’s consolidation has no basis on what your dyspraxic/arthritic/actually-rather-girthy hand sausages jammed into the keys. That’s a motor skills thing. It’s based on what your mind put out in response to the challenge.
For example: for 上る I put in “climb”, rather than “to climb”. I knew the climb that I meant was the actual act of climbing (like you do with a ladder or hill), and so the memory I was consolidating was a correct one when I re-entered it as “to climb”. You’ll have to forgive the simple example: Level 3 and all.
If we count ourselves as wrong when we are factually correct, by not using the ignore/retry/etc script at all, we’re just wasting time (although we aren’t damaging our learning system). The link between the SRS and our memory has been broken, but will become in sync again when the item rolls to where it was before.
Getting an item wrong and having to relearn it isn't a punishment.
It is simply what has to happen to ensure your memory of it is 1) correct and 2) consolidated as much as possible. The SRS doesn’t care how much you’re getting correct, it’s simply doing what it needs to achieve it’s goal: to consolidate these pieces of information into your long term memory.
To be fair, WK doesn’t make it easier to swallow by making the downgrade red, the colour that leaks out of the user’s wounds. It clearly communicates “you shouldn’t have done that”. When you should have. You should enter items you misremember as what you misremember them as. You should enter items you forgot with “forgot lol”. So the SRS can do it’s job well. So it can teach you the correct answers in the way that works.
Really, the optimal system wouldn’t tell you if you got it right or wrong immediately, as a lag on knowing if you’re correct is supposedly better for consolidation (from Wikipedia, no citation though). Just the list at the end would be better. Perhaps allow the user to look through the incorrect answers explicitly to check for typos, very clear that only typos should be corrected if the user wants to learn efficiently.
If I were abusing the ignore script, I would say uninstalling it and wasting time on typos would be a superior choice to continuing to abuse it.
Because counting typos as wrong wastes some time on a few items, but counting incorrect/forgotten answers as correct makes all the time spent on the item up until that point wasted. And continues to waste time on those items until the user realises that they need to start the item again. And if the user doesn’t realise this, then they’ll be mixing up words because they consolidated misinformation into their long-term memory (or not have the item enter long-term memory at all).