Cards that will not stick

Every deck grows a handful of them. The card you have failed eleven times, that comes back every second day, that you now recognise by the shape of the text rather than by knowing the answer. Anki has a name for these — leeches — and the name is apt: they consume a disproportionate share of your reviews and give almost nothing back.

The instinct is to try harder. The instinct is wrong. A card you keep failing is almost never a memory problem, and more repetitions of a broken card produce a better-memorised broken card.

Diagnose before you do anything

Nearly all of them are one of four things, and the fix differs.

It asks for too much at once. "List the causes of X" with six items on the back cannot be graded — you got four, so is that a pass? Ambiguous grading means your ratings are noise, and noise means the scheduler cannot help you. This is the single most common cause.

It has no context. You memorised the answer without ever understanding the thing, so there is nothing for the fact to attach to. Recognisable in that you can produce the words and could not explain them to anyone.

It is ambiguous. The front could reasonably have several answers and you are being marked against one particular phrasing you happened to write.

It interferes with a neighbour. Two similar cards — two drug names, two dates, two definitions that differ in one word — and you reliably swap them. Each is fine alone; together they are a coin flip.

The fixes, in order of how often they work

Split it. A six-item card becomes six cards, or a card per pair. This feels like making more work and is the opposite: six unambiguous cards are reviewed correctly and leave, while one ambiguous card comes back forever.

Add a cue to the front. Not the answer — a discriminator. If you keep confusing two similar cards, put the thing that distinguishes them in the question, so you are being asked something answerable.

Rewrite it as a question you could actually be asked. Cards that echo the phrasing of your source tend to test recognition of that phrasing. Ask it the way an examiner would.

Go and learn the underlying thing. If the card has no context, no rewrite will save it. Read the section again, then rewrite the card from understanding. This is slower and it is often the only thing that works.

Delete it. Genuinely. Some facts are not worth their review cost, and a card you have failed eleven times has already told you its price. Deleting a card is a legitimate scheduling decision, not a failure of character.

What is not on this list: reviewing it more often, adding a mnemonic to a card you do not understand, or simply pressing hard every time and hoping.

What repeated failures cost you

Worth knowing, because it explains why leaving them alone is expensive.

In most spaced-repetition systems each card carries an ease multiplier that decides how fast its intervals grow. Failing a card cuts that multiplier, and it does not come back quickly. A card failed repeatedly ends up pinned near the floor, which means that even once you finally learn it, it keeps returning far more often than it needs to. You are paying interest on a debt from three weeks ago.

That is the real argument for fixing or deleting rather than grinding: the grinding leaves damage behind even when it eventually works.

In Anki specifically, this is what leech handling is for — after a set number of lapses it tags the card and suspends it, which is a forced interruption to make you go and look at it. If you use Anki, leave that feature on. It is doing the thing this page is asking you to do manually.

How Quat handles it — and what it does not have

Quat is an iPhone and iPad app I make, so read this as an interested party's description, including the unflattering part.

The scheduler behaves as described above. A card rated again returns in the same session and loses ease; ease has a floor, so a repeatedly failed card settles into short intervals. Cards are editable at any time, which is the important one here — splitting or rewriting a bad card is the actual fix, and you can do it from the Cards tab without ceremony.

What Quat does not have: leech detection, suspend, or bury. Nothing counts your lapses, nothing tags a problem card, and nothing will interrupt you to say this one has cost you eleven reviews. So on Quat the diagnosis above is entirely manual — you notice the card, or you do not.

If your decks are large enough that leeches are a real tax, Anki's automatic handling is better than anything Quat currently does, and that is a straight comparison rather than a hedge. What Quat gives you instead is the stamp on every card saying where it came from, which makes the rewrite fix practical: you can go back to the lecture or the page and write the card properly, instead of guessing at what you originally meant.