How to force FSRS to show today's (re-)learned cards tomorrow (1d interval)

Fine. I just worry that if I do not respond, users might assume that I implicitly endorse this practice, especially when I have involved this discussion.

Here I want to explain why I do not recommend it.

The 1d+ learning step is preferable as it only affects the first inter-day interval, allowing FSRS to still accurately calculate memory stability based on this interval. However, if the parameters are adjusted, the subsequent calculations of memory stability will all be inaccurate.


There is no one here who would recommend such a thing. But endlessly dissuading the user from what he wants seems rude to me.

Here I want to explain why I do not recommend it.

Out of curiosity I did tweak with the parameters both in both links you provided. The solution seems* to be to set the initial stability parameter for Good to that of hard. It does affect future intervals, but not if you ignore the first step. The first step is “wasted,” beyond that the graph seems* to stay unchanged. This seems* to eliminates the problem of having wacky reversed interval times for Hard vs Good for multi-day learning steps (the one Scherlock mentioned)

*seems, becuse I don’t think I understand it enough to say anything definitively

I would still rather stick with different decks with different desired retentions than to tweak the parameters. But thanks so much all for your inputs.

If we make the warning too long, it’s going to ignore users who want to ignore it.

What about recording it in Getting help - Frequently Asked Questions (

By that, do you mean adding a separate FAQ entry to that website? It might be better done by mentioning it in the help text for the learning steps section, or on the section of the manual that is linked to.

I guess you mean here: Deck Options - Anki Manual (

That’s the latter, yep.

Do you actually have problem answering the card in 3-4 days?
For me FSRS typically puts a new “Good” card after 9-11 days, sounds like a lot, but if you remembered, you remembered, I don’t see why invent artificial constraints for this.
Also important to remember that Anki is primarily a review tool, for material that you learn (pick apart, try to understand the “why” behind the answer) before/during card creation.

Did you add the points Sherlock mentioned after that?

He mentioned three things,

  • Hard interval might be longer than Good/Easy or that Easy interval might be shorter than Hard/Good.

  • desired retention function may not work properly.

  • For 1+ days of (re)learning steps, the new interval given by FSRS might be shorter than the previous interval.

I believe the 2nd point is really crucial for us to know.

(Re)learning steps of 1+ days are not recommended when using FSRS. The main reason they were popular with the old SM-2 scheduler is because repeatedly failing a card after it has graduated from the learning phase could reduce its ease a lot, leading to what some people called “ease hell”. This is not a problem that FSRS suffers from. By keeping your learning steps under a day, you will allow FSRS to schedule cards at times it has calculated are optimum for your material and memory. Another reason not to use longer learning steps is because FSRS may end up scheduling the first review for a shorter time than your last learning step, leading to the Hard button showing a longer time than Good.

I did read that but the second point isn’t clear at all. At least for me, I wouldn’t have known that desired retention may not work properly if you use learning steps of 1+ days unless Sherlock mentioned that.

Is that what you mean?
Indeed, Sherlock did not explain what this means. Perhaps that is why it has not been added to the manual. I assume that due to the fact that the training step can be set more than the FSRS (If this learning step were missing) would assign an interval, the true retention may become lower than the “desired retention”.
For example, you will assign a training step of one year. But after 4 days, FSRS considers that card retention = “desired retention”. And the longer you can’t do a review due to a large learning step, the lower your “true retention” will be.
That is, a big learning step does not allow you to do reviews on time, so it may happen that “desired retention” <“true retention”.

If so, it seems to me that this is already enough in the manual:

By keeping your learning steps under a day, you will allow FSRS to schedule cards at times it has calculated are optimum for your material and memory.

This can very easily be interpreted as, and this is how I did it, that by keeping your learning steps under a day you allow FSRS to take over the scheduling the very day you start learning something. This is probably because a similar point is raised by some other people (or maybe this is the place where I found this idea, idk).