FSRS does not respect manually scheduled intervals

[hnvy]

Introduction

Hi all,

It seems that FSRS does not respect manually scheduled intervals.

For example:

1. Create a new card
2. Manually schedule it (Ctrl+Shift+D) to 190! (this will set the interval to 190 days)
3. Review it after 190 days
4. FSRS will show that the next review time is 2-4 days (or similar)

Is this a bug? Or an intended behaviour?

If it is an intended behaviour, how would one overcome it?

Expected outcome

I expect that the interval of 190 days would be recognised by FSRS. If I remember a fact after 190 days, it should not matter how I came to that point (e.g., naturally vs. manually).

At least these were my observations from using SuperMemo 18 and SuperMemo 2.

Videos

See the videos below . I have tested it multiple times with different methods; the outcome was the same:

1. Changing the date and time of my device
2. Using filtered decks
3. Grade the card and introduce it to the learning process naturally, and then manually reschedule it

SuperMemo 2

FSRS

[Keks]

When building the next interval, FSRS does not use the previous one.
This issue was discussed somewhere on github anki.

[Expertium]

@L.M.Sherlock

[hnvy]

I searched online, but only came across this one about an addon: [Question] Compatibility with Add-on "Learn now & Grade now" (Documentation) · Issue #540 · open-spaced-repetition/fsrs4anki · GitHub. I will look around for more later tonight.

My question would then be: How would I introduce a bunch of cards to the learning process without having to go through all of them naturally?

[Keks]

It was mentioned here Assigned memory state after 'set due date'+optimize incorrect · Issue #2998 · ankitects/anki · GitHub

[Keks]

@hnvy

My question would then be: How would I introduce a bunch of cards to the learning process without having to go through all of them naturally?

Give more information about what you are trying to do and why.

[hnvy]

Let’s say I have memorised a chapter in a textbook.

Then, I add 100 cards.

When adding cards about a topic, I have already memorised the content. So, the first few natural steps are quite useless to me.

So, in SuperMemo 18 (and Anki’s SM2), I would execute a manual review and spread the content over a few days. This would save me 200 unnecessary natural reviews for today, and an unnecessary 100 reviews tomorrow!

But anyway, I think I have figured a (partial) way around it!

1. Create a new card
2. Grade it naturally
3. After the above, manually schedule it (Ctrl+Shift+D) to 190! (or whatever)
4. Review it after 190 days
5. FSRS will now respect the interval, or so it seems (because there will be a Δt FSRS explained, part 1: What it is and how it works : Anki ?)

My students suffer from the same issue too; we will wait for the integration of the Grade Now addon into Anki

[Danika_Dakika]

This thread includes links to some of the other places this issue has been discussed (Forums, Discord, Github).

FSRS with known new cards

This is the best solution that they’ve been able to find for FSRS, because there’s no way to know if the interval created by Set Due Date is “real” or not. [It was touched on briefly here, but the longer conversation is on Discord, and a few different places in GitHub.] When you’re using it to scatter a set of cards across the next few years, you’re not using any criteria to determine the interval. So you don’t necessarily know the card set for 500d better than the card set for 100d. FSRS can’t…

[Keks]

If you want to rate all the maps without looking at them, use Auto Advance.
If you really always remember these cards, FSRS can increase the initial stability up to 100 days during optimization.
The initial stability can be changed manually, then it is not recommended.

[hnvy]

@Danika_Dakika Thank you for the link! I have worded my question quite badly, and hence none of this came up during my search !

If you want to rate all the maps without looking at them, use Auto Advance.

We have explored this option already. We would probably need to create a filtered deck with the cards of interest so as not to accidently run it on Review cards.

The initial stability can be changed manually, then it is not recommended.

Ah! I am guessing through manually changing the database…?

[Keks]

@hnvy

Ah! I am guessing through manually changing the database…?

By changing the FSRS parameters in the preset.
A technical explanation of the FSRS algorithm

But the best thing you can do is to honestly evaluate your cards. And if your Pre-learning is as good as you think it is, then when optimizing FSRS, you will get large initial intervals.

[sorata]

I believe there was only one person who used ‘Set due date’ to equally distribute cards over a period of days. He could’ve done this using filtered decks, but that one change he suggested seems to be affecting a lot of people. If FSRS is to become the default scheduler, those changes should be reverted. I am hoping this is on Sherlock’s To Do list.

(Actually, isn’t this workflow unfeasible using SM2?)

[sorata]

Let me explain what Keks is trying to say. If you consistently memorise your new cards outside of Anki, FSRS will recognise that the stability of your newly learnt cards are higher than, say a few days. Thus, if you can usually recall 90% of newly learnt material 10 days after you learned it, your initial intervals will be that much high.

The point to note here is that FSRS is already trying to save you from those “unnecessary” initial reviews (except the initial learning steps which you can just set to 1m).

One more thing to note, is that there might be cards in your deck you don’t learn outside Anki. If this is the case, FSRS will have a hard time generalising for your cards. One trick might be to graduate such cards with the Good button whereas graduate the other type of cards with the Easy button. If I’m not wrong, this should give you lower intervals for one set of cards, and higher intervals for the other.

You can also use ‘Ignore reviews before’ option if you have a lot of old reviews you don’t want FSRS to optimise with.

@hnvy

But anyway, I think I have figured a (partial) way around it!

Thanks for this! I didn’t know about such workarounds.

[Expertium]

You can also use ‘Ignore reviews before’ option if you have a lot of old reviews you don’t want FSRS to optimise with.

It ignores card, not reviews, which is why it will be renamed to “Ignore cards reviewed before”.

[sorata]

Let us not overwhelm them with information

[hnvy]

Thank you for this explanation. Much clearer now!

@sorata

One trick might be to graduate such cards with the Good button whereas graduate the other type of cards with the Easy button. If I’m not wrong, this should give you lower intervals for one set of cards, and higher intervals for the other.

I have been thinking about doing this (this is my workflow in SuperMemo 18 and SuperMemo 2). But, I was worried that this kind of behaviour would affect the way FSRS views my collection as a whole.

I will give it a go, and see.