Question About Handling New Cards

Hello everyone,

I’ve been using Anki (mostly on Android) for a year now, and I have a question. When you see a new card for the first time, you’re supposed to be discovering it, so you wouldn’t know the answer yet. Do you always press the red “again” button for each card’s first appearance? Is this how you handle new cards as well?

If I don’t know it I use Again yes. Not everytime though. Sometimes I do know the answer if I’m studying for school. Any particular reason you are asking this now after a year?

For a long time, I used to answer “correct,” but this recently caused issues with the new FSRS algorithm. I just want to make sure I’m using Anki in the best way possible. Thanks for your response!

That’s not necessarily the case. For instance #1 of the Twenty Rules is, “Do not learn if you do not understand.” At its heart, spaced repetition is about memorizing things you already know about, so you wouldn’t be discovering information for the first time on a flashcard.

But, as you might imagine, the situation might be different depending on your subject of study, your reason for studying, whether you made the cards yourself, etc. Maybe at one end of the spectrum you could put – “studying for a class, and wrote every note/card myself” – at the other end – “recreational learner, and using premade decks” – with many permutations in between.

Regardless of where you fall on those spectra, your priority should be to grade your answers honestly and accurately for every card. If you are a language-learner with words in your pre-made deck you’ve never even thought about – yes, you should be grading those Again the first time you see them (and maybe even the 2nd-3rd-4th…). But if this is a word you just created a note for last week after stumbling upon it and studying it a bit – maybe it’s not all that new to you, and another grade would be more correct.

They are mostly words and phrases in a foreign language.

Not knowing them is normal. I don’t learn everything in order I create them. I get my frequency dictionaries from here and here then reposition everything with the frequency. This allows me to learn the most frequent words first which is also the easier way of going about learning vocabulary.

As for FSRS, it generally gives you longer intervals initially. The reason is FSRS is calculating the intervals from the Stability (S) of cards unlike SM2 where you have a pre-determined graduating interval for every card. This is normal and you shouldn’t need to worry about. It is also recommended that you don’t change your grading habits. Please read these FAQs.

This might be relevant:

Q6: My first interval is too long! Is this normal?

A6: In short, giving long first intervals is one of the strengths of FSRS. Don’t be surprised if your first interval for “Good” is close to a week and your first interval for “Easy” is several weeks long. Read further for a deeper explanation:

For many users, the default algorithm (SM-2) tends to show new cards at unnecessarily short intervals. So, when users switch to FSRS, they tend to feel that the intervals given to new cards are too large. But these larger intervals match the desired retention better. By using these larger intervals, FSRS can prevent many of the unnecessary reviews that happen when using SM-2. So, it is advisable to try using these larger first intervals for a few days and see how it goes. It’s worth mentioning that for mature cards, the opposite is true: FSRS is more conservative than SM-2.

If you still want to decrease the intervals, you can increase your desired retention. But note that this will decrease all the intervals, not just the first intervals.

The ideal time to review a card is when you’re just starting to forget the information but can still recall it. If FSRS sets intervals that are too long, I’m not sure it’s a strength if the information fades and you have to press “again” frequently. This algorithm can be quite unsettling, but I think I’ve managed to tame it by tweaking the settings.

You’re right. Therefore, FSRS by default shows you cards after the interval when you’ve already forgotten 10% of the information. You can adjust that setting and make FSRS work better for your purposes. It’s possible to make FSRS show you cards when you remember more of the material. SM2 on the other hand, by default, shows you cards at unoptimal intervals because everything is forced to have 1d graduating interval. Remember that it is not possible to know when exactly you’re going to forget a card, at least not until you give Anki access to your brain. And we don’t even know where memory is located in brain (please don’t say in synapses between neurons, I’m pretty sure nothings there). FSRS is as best as it gets. Also, AFAIK forgetting curve is steeper initially and then it flattens so I’m not sure what “you’re just starting to forget” mean unless you’re talking of individual cards, which no scheduler can’t say when will be forgotten.

This algorithm can be quite unsettling, but I think I’ve managed to tame it by tweaking the settings.

Can you tell me exactly what you tweaked? The only setting that you have to care about is desired retention. FSRS parameters are taken care of by the optimizer, you don’t have to tweak them manually.

Anyway, some theory. As the probability of recall drops, the increase in memory stability in the event of a successful recall increases. In other words, reviewing a card when the probability of recall is low produces stronger memories. However, this doesn’t mean that the best schedule is when all cards have a 1% chance of being recalled. It would not only be demoralizing (I mean, imagine only getting 1 card out of 100 right), but also inefficient as you would spend too much time re-learning forgotten material. The optimal desired retention seems to be between 70% and 97% for most people.


Apart from what Expertium said, I will also reiterate this. The probabilities calculated here will only make sense if you talk about a large number of reviews. Like coin tosses. There’s no way to know when exactly you’ll be unable to remember a particular card. Even if you connect your brain to Anki, and here’s our understanding becomes a bit messy, you can still be unable to remember something. Even if FSRS tries to predict when exactly a person forgets something, it will fail. Using probabilities makes more sense here. To give you an example, has it ever happened to you that you couldn’t remember something for a while, but you suddenly recalled that thing after some hours. Worst case scenario, that has happened in a test.

One more thing, you’ll need a lot more effort and time when you’re “at the brink of forgetting” all your cards. So that might not be a good idea for your learning rate.

I restarted the optimizer after several weeks, hoping that the algorithm would see my failures and adjust future intervals accordingly. Attached is a typical example of a card that I forget 50% of the time because FSRS suddenly switches to a much larger interval.

It seems that you are using Hard instead of Again because there are no Again entries in the card info, only Hard entries.

From the FSRS tutorial:

FSRS can adapt to almost any habit, except for one habit: pressing “Hard” instead of “Again” when you forget the information. When you press “Hard”, FSRS assumes you have recalled the information correctly (though with hesitation and a lot of mental effort). If you press “Hard” when you have failed to recall the information, the intervals will be unreasonably high (for all the ratings). So, if you have this habit, please change it and use “Again” when you forget the information.

No, the only time I chose “hard” was because I found the answer but it took more effort than usual, so the “hard” button was appropriate. However, at the very beginning, I marked “correct” even though I had just discovered the card (Basque vocabulary) for the first time. In that case, I probably should have chosen “again.”

I think that doesn’t matter as long as you don’t know any of those cards. If you know some of those cards then it might be a problem as you’re rating everything Good.

That is certainly an issue. Are you reaching your desired retention though? Can you share the numbers from the bar graphs from statistics that show retention?


Go lower.

Is this ankidroid? If yes, take such screenshots in the horizontal position of the phone for additional information. And if possible, change the language of the application into English during the demonstration.

You have remembered the card 6 times. It doesn’t look like forgetting 50% of the time.

It looks like you haven’t completed the card training in one day.

1 Like

People want you to do the translation :smiling_face_with_tear:

To see your retention, look for Stats > Answer buttons > Mature % correct. How long have you been using FSRS? It can be useful to compare a short timeframe (when you’ve been using FSRS) to a longer one (that includes time on the default SM-2 algorithm).

To talk about how FSRS is interacting with a particular card, we’d need to see the full Card Info (top to bottom) – in landscape, if you need to get it from a small-screen device (as Keks mentioned), so we get all 6 columns and timestamps. It would also help to know your learning/relearning steps, and your FSRS settings (with the parameters as text, please).

Trying to compare SM2 and FSRS retention might be misleading.