And then... I hit this mathematical limit where I just couldn't keep up. You can't add 20 new cards a day for any length of time without the review cycle getting overwhelming. I'm not sure it's even physically possible to Anki full recall of everything learned in a college course - each class would need to be much longer than a term to keep up with the review.
One solution is to just keep plugging away, reviewing what you can and hoping the things you answer correctly will relieve the pressure as they come up less frequently. But since you're late on many card, recall sucks and I never really made much headway.
I also have a really hard time retiring cards. It's like deciding to cross things off your TODO list that you'll never do, you always want to keep it on thinking you might do it someday. Same with a not very relevant piece of info, it's right in front of you, it still feels like it might be useful to know...
1. You have to take breaks from adding cards, and just be in review mode sometimes.
2. You have to tweak the settings. The defaults are bad. For instance, you probably want to adjust the percentage by which a lapsed card's interval shortens. If you lapse on a ten month card, you don't want it resetting to just several days, but say to 50%: five months. Another thing is to tweak the custom intervals to reduce the number of initial reviews for newly added cards, and also tweak the interval steps for lapsed cards to control how many times a card is reviewed before being returned into the deck.
I've thought about this quite a bit and written about strategies and tips here: https://senrigan.io/blog/everything-i-know-strategies-tips-a...
A large part is also now using the proper visualization techniques (ie. using intensive memories, an anecdotal use case is I've memorized certain ideas behind neural networks by always putting them in a "closet"). This helps. Truncation of dumb cards is useful.
Another really important thing is just to delete an entire deck (which I've done), when I realized I frankly ... just don't care about the subject any more (in this case, music composition).
Using Anki is very helpful if you need to repeat much information for a small period of time, e.g. before exams or interviews.
I might go for a couple of days without adding anything, do a couple days adding 5, add 10, 15, 20, or even 30 depending on the circumstances.
Correct. However, using a huge premade deck is not a bad idea in the sense that it's a huge time saver.
For me, creating an anki card takes between 1-2 minutes, depending on how slow the dictionaries (plural) are. So, 30-60 words per minute (45 avg) means it would take me almost 4.5 hours to input a measly 200 words.
I can write down 200 new words from one episode of a tv show. You can see where this is going. For Korean, about 5-6k words are recommended to begin having non-basic discussions. 5k words at 45 words/hour is ~110 hours.
Instead, I suggest using a parent deck that has a premade deck + your personal deck as children. This lets you have the best of both systems. You can use the Hoochie Mama! plugin  to properly randomize your review queue for child decks. Anki by default pulls cards in order from child decks.
Note: All of this assumes that your pre-made deck is of good quality. I can't speak about other languages, but Evita's Korean deck (5.8k cards) is sorted by frequency. It's fairly good, at least until you get to more of a intermediate level.
What I ended up doing was modifying a pre-made deck to have more fields, and I edit a card as it pops up into my new queue. This takes more time, but it makes it easier for me to learn a word seeing example sentences and other information.
I think there's also a setting to limit new cards based on review size, and that also helps avoid review hell. I imagine it's not the default because a popular use case is students in courses where there are high stakes deadlines.
There are a few big issues, but the easiest one to fix is the reset interval.
When you forget a word that you're reviewing (that is, when you hit "Again"), anki resets the interval to 0%, as if it were a totally new card.
This is bad for 2 reasons. 1) A lapse is not the same as learning a completely new word. When you review a lapsed card you're just refreshing that connection that you've already made. So it's wasteful to completely reset the interval unless you've really totally forgotten the word.
2) The interval reset can be huge (going from 7 months to 1 day) so it makes it tempting to just hit "Hard" instead. I know I've done that.
The fix here is to set a more reasonable reset value like 40-60%. I use 45%.
The other big issue is the algorithm works differently than you would expect. I discuss this in more detail in my post, but essentially anki gives each card a modifier value. Starting at 250%. Every time you hit Again/Hard/Easy, it -- permanently -- modifies this card. This is quite unintuitive! When people grade a card, they're really grading their own memory / ability to recall. Not how hard the card itself is.
Why is this bad? Because if you hit "Hard" a lot, then you're seriously capping the interval growth rate for a word. In other words you'll see it way more often than you need to. Most people don't hit "hard", -then- "easy" once they learn it. A lot of people I know hit "hard" to be on the safe side instead of "good", purely out of habit. This picture explains why a low modifier is bad. 
The way I fix this is by using 2 plugins: the first makes grading buttons not modify a card's modifier, and the second one resets ALL cards to 250%. Low-Key anki goes over this in detail. 
However! This fix assumes a relatively homogeneous deck in terms of card difficulty. I.e. words for a language. If you're studying, say, algorithms or something where the difficulty may be heterogeneous, then you may want anki's default behavior.
I have been using Anki for Chinese with a premade deck (Spoonfed Chinese, if anyone is curious) for around 1.5 years and reviewing time has crept up too much. I have lowered the number of new cards per day several times (started at 20, now at 12, thinking about going to 10) but I am very averse to change the rest of the defaults since they are very technical things and it's hard for a layperson to know the optimal settings. I trust the Anki devs to have more data on, e.g., the typical rate at which we forget things than I, a random language learner, can have; what do I know about that? So I fear that changing something may harm my learning (currently I get more work than I would like, but I do learn).
You seem to know what you're talking about so I'll try your advice (thanks a lot for it!). But if any Anki dev is reading, please, pay attention to the defaults. I'm sure the overwhelming majority of users are going to use them, out of lack of information and/or fear of tweaking things that we don't know much about. It's a pity to have such an awesome software not live up to its potential due to bad defaults.
Would it be hard to ask the user a small questionnaire, like "how long are you planning to study this subject?", "do you have a specific goal (learning X cards by date Y)?" and "what is the average and maximum time you are willing to devote per day?", and use that to set reasonable options for that user?
IIRC the anki defaults seem to be geared more for short term, dense studying (like for tests). In those situations you would want a lapse to reset the interval to 0.
For longer term learning, you ideally want an 80-90% recall rate.  In my post on anki I have two studies that look at the effects of longer intervals on learning. 
If you hit "hard" on a card, and then months later it re-appears, at which point you now remember it well and hit "good", doesn't it reset the state of that card back to normal?
This is why people who use Anki long term recommend not hitting the Easy button: it’s a permanent +15% modifier.
This algorithm -does- show you certain cards sooner or later but it does it in a nonintuitive way.
Cards marked “Hard” like this will end up being seen way too often. This picture (1) explains the problem.
Everybody I would urge you to read this comment if you are an Anki user.
The point about Anki is not the sw but the intervals and if you haven't got them right, there's not much point.
The real key is to have a long commute every day where you get to review on the train. My commute got shorter this year haha
Sure, it's not optimal to miss a scheduled recall because then you might forget the card, which is quite bad. But most of the time you'll be doing OK, so it's always worth trying more cards.
I find SRS software like Anki is best taken with a "slow and steady wins the race" approach. I only do new cards when I start running out of daily reviews now, and I find it a lot less punishing.
If you keep on failing to memorize a card, then you should suspend or delete it. I know this sounds crazy, but you end up wasting so much energy on those cards; energy that you could have spent on memorizing characters that you don't struggle with. When you do anki for years this really adds up.
If that character is "important" relative to other characters, you can take it outside of anki and write like 15 sentences with it, etc.
You can filter by suspended (or leeched, I think) cards easily. What I do is set aside time once every ~2 weeks to go through my suspended cards and see if they need to be modified to be simpler, etc.
What really kills you is missing reviews. I did kanji study at 10-15 words/day for a few months, and then missed reviews due to work/life issues. Going back and slogging through that backlog of ~400 reviews was very demotivating.
I recommend changing the defaults of mixing new cards and reviews, to showing reviews first. That way it's easier to focus on knocking out reviews, which only really takes like ~15-30 minutes. Even at 20/cards day.
I have found in the past that
(1) with sheer persistence, I busted through leeches. For instance 催促 used to be a leech, in a deck of cards where that is the front. Now I know the word so well I can recall it in the reverse direction: for instance to talk about how it used to be a leech. :)
(2) given enough time, cards you know well will turn into leeches, simply due to occasional gaffes. The conversion of a card to leech is based on an absolute count of how many times you got it wrong, ever, regardless of how many times in between you got it right.
Thats totally fine too, it’s just that energy could’ve been dedicated to other cards where there isn’t an issue. I also power through some of these occasionally if I don’t feel like busting out pen and paper, etc.
> given enough time, cards you know well will turn into leeches,
Yeah this is an issue. Particularly if you keep periodically doing Anki, with just enough time to have lapses. So I set Anki to mark leeches instead of suspending them, and when I go through suspended / leech cards I do an assessment.
Nope! It doesn't work that way; each card is scheduled independently of others. You're not going to see other cards less due to seeing one card more.
Busting through leeches just adds to the workload; you still spend all the same time on the other cards.
(Even if you set daily limits on reviews, that just causes a backlog.)
It's a proprietary, closed-source program for Windows that runs as a web service on localhost, using your browser for UI.
It's specialized for learning Kanji meanings and readings, nothing else. You can't use it for Japanese vocab, you can't use it for Russian, or for memorizing capital cities or dog breeds.
It does one thing, and well.
> Going back and slogging through that backlog of ~400 reviews was very demotivating.
You will love Tankan if you try it, because in Tankan, you can easily test yourself on 400 Kanji in the space of some 15 minutes, and refresh your knowledge efficiently.
Tankan is not based on self-assessment ("do I know this card?"). It generates test sheets where you fill in answers for each Kanji. The machine decides if your answers are right or wrong, not you. The Kanji are all displayed in one page. You can read the next few kanji ahead while your fingers are answering the current one. You can tab quickly from one field to the next. If you don't remember something, you can skip it, and then backfill it before submitting the test for grading.
When the test is graded, you can then instantly take another test which just has the kanji that you got wrong. You can iterate on this until everything is right, and the the next batch of cards will be drawn from your deck.
When you know the material well, literally your typing speed is the critical factor. The problem with flashcard programs like Anki is that you're blocked from seeing the next card while on the current card. And even if you aren't, you're limited to seeing one at a time. Your brain needs to see a "sliding window" of multiple cards so it can think ahead.
For learning, Tankan has a rich selection method for kanji; by grade, frequency, stroke count, string search.
Large sets of kanji can be partitioned into sets labeled 0 through 9, A through F, which are selectable through checkboxes. So, say you want to learn some set of kanji according to some selection criteria, which happen to produce a set of 1000 kanji. What you can do is use the 16 subsets to divide it arbitrarily. Say, make a four week plan: This week you drill on sets 0+1 early in the week and 2+3 later in the week. Then the next week, 4+5 and 6+7, the over the remaining two weeks 8+9 and A+B, and C+D and E+F. Then, you start combining them.
Before long, you can casually take a test of all 1000, in one sitting. Bang away at the keyboard for half an hour. Have it graded. 92%, very good! That means 920 were right; only 80 are wrong: just a couple of minutes to re-test on those and after that round, maybe 7 will have to be done again in just seconds.
Tankan has a natural vacation mode: when you've drilled yourself, just put it away. Come back to it in weeks or months, fire up those 1000 kanji, slog through it, and you're back in the game before you know it.
Tankan also identifies which kanji you're confusing for one another. You can select those as a deck, and drill yourself on it. In this mode, the confused groups of kanji always appear together in the test. The groups themselves are mixed up relative to other groups, and the order is scrambled within the groups, but the confused kanji are always in successive runs. The relation is transitive. If you confuse A for B, and also C for B, then A-B-C are a confused set. You will be startled! If you're mixing up 住 and 注 主, it will know. If you're mixing up 姉 and 妹, it will know.
Tankan keeps no permanent record. Only the most 20 recent answers for any kanji are retained, and are not associated with any scheduled time, or time interval or anything. You must click a save button to explicitly checkpoint your state. When the program starts up, it doesn't automatically load the saved session state; if you want, you can ignore it and start in the initial game. It works exactly like checkpointing a game: re-start the saved game, or begin a new one.
There is none of that "Anki tyranny" ("OMG, my 3.5 year card lapsed to 1.0 years, waah!"). It doesn't take over your life in any way.
> There is none of that "Anki tyranny"
Anki isn't really tyrannical, it just has some defaults that aren't good in my opinion. If you lapse on a word, it's likely that the interval -should- be reduced a bit.
I think the idea is that eventually you'll be recalling even those "leech" items, due to the sheer exposure you're getting in the meantime. Frustration is part of the game.
This helps less than you'd think. Many of my leeches are due to word similarities. Sometimes I'd have an issue of erroneously swapping two similar words, and my brain would get "stuck" following that circuit. If that makes sense?
The fix that I found was suspending one or both of the words and making up a lot of sentences for them. And physically writing them.
When this happens, I find the Anki "leech" feature to be useful. I just let it remove (suspend is the word, I guess?) those words automatically. Flashcards are not my sole learning method, so I will probably learn them anyway by other means at some point, so spending lots of time on repetitions to get them right is not worth it.
8 strikes and you don't have to learn it!