← seansooch.devField Toolkit
Force Multiplier 03 · Memory

Anki

A flashcard program that decides when to ask, so each fact comes back just before you would have lost it.

Bottom line

Two findings carry the whole method: pulling a fact out of memory strengthens it more than reading it again, and reviews spread out over time beat the same reviews packed together. Anki does the bookkeeping for both. Your job is to write small cards and show up every day.

1 · In plain wordsWatering on a schedule

A new seedling needs water every day. An established plant needs it once a week, and a tree hardly at all. Water everything daily and you waste the day. Water everything monthly and the seedlings die.

Facts in memory behave the same way. A fact you met this morning needs to come back tomorrow. One you have recalled correctly five times can wait months. Anki keeps a separate schedule for every card, so the fragile ones return soon and the sturdy ones stay out of your way.

That is why a deck of several thousand cards is workable. On any one day you see only the small share that is about to fade.

2 · The mechanismThree numbers per card

You are shown a question. You answer from memory, reveal the card, and grade yourself: Again, Hard, Good or Easy. The scheduler turns that grade into the next due date. Current versions of Anki ship a scheduler called FSRS, which tracks three numbers for every card.

Retrievability
The chance you could recall the card right now. It starts near 100% after a review and falls every day.
Stability
How slowly it falls: the number of days until that chance drops to 90%. A good review raises it, so the next gap is longer.
Difficulty
How stubborn this card is for you, on a scale of 1 to 10. Hard cards gain stability more slowly.

You set one dial, called desired retention: how likely you want to be to remember a card at the moment it comes due. The default is 90%. The scheduler then waits until a card's retrievability has fallen to that line, and asks.

Try the dial

A forgetting curve: the chance of recall falls over days, and the review is due where it crosses the desired retention line. 100% 75% 50% 0 DAYS SINCE THE LAST REVIEW
10 daysuntil this card is due.
Figure 1. The forgetting curve FSRS uses, in its published FSRS-4.5 and FSRS-5 form. The newest version also fits the steepness of the curve to your own review history. Move the dials: asking for more retention pulls the due date in.

The same relationship as a table. At 90% the gap equals the card's stability. Everything else is a multiple of it.

80%
Wait 2.40 times the stability. Fewer reviews, more forgetting.
85%
Wait 1.64 times the stability.
90%
Wait exactly the stability. The default.
95%
Wait 0.46 times the stability: every gap is less than half as long.
97%
Wait 0.27 times the stability: every gap is about a quarter as long.

This is why the Anki manual warns that above 90% the workload rises very quickly, and that above 97% it can be overwhelming. The last few points of retention are the most expensive ones.

3 · Set it upSix settings and habits

  1. Turn FSRS on. It is at the bottom of the deck options page, and it is switched on for the whole collection, not per deck.
  2. Leave desired retention at 90% unless you have a reason. Raise it a little for an exam you cannot afford to miss, and expect more reviews.
  3. Press Optimize about once a month. It fits the scheduler to your own review history. The manual says not to edit the parameters by hand or copy someone else's.
  4. Keep learning steps short. The manual advises against steps of a day or longer when FSRS is on, and against having many of them.
  5. Do reviews before new cards. Every new card is a promise of future reviews. If the pile is growing, add fewer new cards. Do not skip the reviews.
  6. Grade honestly. The schedule is built from your grades. Pressing Good on a card you missed tells the scheduler to wait longer for a fact you do not have.

The desktop program is free, and so are AnkiWeb sync and the Android app. The iPhone app is paid.

4 · The half the software cannot doWriting cards that survive

One fact
One card asks for one thing. If the answer has three parts, it is three cards.
Understand first
Make the card after you understand the idea. A card is for keeping something, not for meeting it.
Cloze
A sentence with one piece blanked out is the fastest way to turn a line of notes into a card.
Pictures
For anatomy, films and diagrams, hide a label on the image and ask for it. Anki has this built in as image occlusion.
Your words
A shared deck saves weeks of typing. Reword any card you keep missing so it asks the question the way you think about it.

A deck is a database

An Anki deck exports as a single file, and inside it is an ordinary database of notes, fields and tags. That is what makes it a multiplier beyond flashcards: the same facts can be read by a program and turned into something else.

The study site linked from this portfolio was built that way: seven question banks derived from Anki decks, 12,357 guide facts, and 12,972 board-style questions written from them. Building it surfaced about 100 factual errors in the source decks, which were corrected. Treat any deck, shared or your own, as a draft that can be wrong.

5 · What most people get wrongFive traps

  1. Making cards and calling it studying. Building a deck feels productive. The learning is in the reviews.
  2. Cards that hold a paragraph. One grade cannot describe six memories. You pass the card while losing half of it, or fail it for one slip and see the whole thing far too often.
  3. Chasing 99%. Past the default, each extra point of retention costs more reviews than the last. The time is usually better spent on new material or practice questions.
  4. Knowing the card and not the fact. After enough repetitions you can recognise a card by its shape and answer without thinking. A fact you can only produce for its own card is not much use on a ward or an exam. Pair the deck with practice questions that ask for the same fact a different way.
  5. Skipping days, then quitting. Reviews you skip do not go away. They arrive together. A short session every day beats a long one on Sunday.

6 · Check yourselfThree questions

RecallIn FSRS, what does a card's stability measure?
Show the answer

Days until recall falls to 90%. The 1 to 10 number is difficulty, and the review count is not one of the three numbers at all: two cards reviewed five times each can have very different stability.

Apply itYour exam is in six weeks and reviews are taking longer every day. What do you change?
Show the answer

Fewer new cards. New cards are what feed the review pile, so that is the dial that brings it down. Raising retention to 97% shortens every gap to about a quarter and makes the pile worse. Skipping moves the same reviews to one bad day.

What would breakYou put a whole table of eight values on one card. What breaks?
Show the answer

The grade. The scheduler keeps one stability for the card, so it cannot bring back the two values you missed without also bringing back the six you know. Eight small cards each get their own schedule.

SourcesWhere this comes from

More Force Multipliers
Field notes, not doctrine. Check the linked sources before you rely on them.Written with Claude Opus 5.5 · All Force Multipliers