I personally use for some deck setting, when I leave re-learning steps empty. If I press Again on a card I want to put it as review for at least the next day - not in learning.
If an update will change that behavior I will need to somehow find a workaround to keep my flow the same.
2 Likes
Me too, but for a different time duration and purpose. But at the same time I want to use the automatic scheduler.
Why does it have to be one thing or the other 
So the experimental version of the <1d scheduler has arrived and I am wondering if it is working correctly.
With a brand-new card, these are the intervals that I have, provided that I have 4 hours local time till my “Next Day starts at” time.
It seems that changing the first parameter to be as low as possible has helped change the again interval to be realistic a bit.
But a first parameter of 0.0001 (it corrected itself to 0.0000) is unattainable through optimization.
I am left wondering if this is the way the <1d scheduler is supposed to work.
But the optimizer cant reach as low as 0.0001 though, am I right 
Yes, but “can” doesn’t mean “will”
1 Like
It has never reached down this low for me at all for all my decks, so it probably never will 
Thanks anyways for the clarification… 
Btw my relearning intervals are 1s 5m 55m 2h 4h.
Here are the intervals for a card I am relearning.
I am getting 28m for a good interval. This is not present at all inside my given set of learning steps.
Is this because of smart fuzz at work, or is the <1d scheduler trying to influence something
Also why is hard 1d but good 28 minutes 
Check with different Desired retention. There may be a bug.
Try changing your Next day starts at so we know what’s going on exactly. Day rollover is obfuscating a lot of information now.
Ok, here I have no idea. Please screenshot your settings
Also @L.M.Sherlock
Beta 3. No learning steps. I get this for cards in relearning:
If I add a single 10 minute relearning step I get this:
FSRS-5 isn’t good at scheduling short-term reviews because its prediction of memory state is not calibrated to short-term memory. I don’t recommend leaving your learning steps field empty if your first four parameters of FSRS is shorter than 1d.
I had an idea.
AFAIK, the reasons for not adding an option to toggle the short-term FSRS scheduler are
- we are not confident about the efficacy
- adding too many options to the UI is confusing and may also induce FOMO.
So, what if we add an option to enable the short-term scheduler through the debug console even if the user has configured the (re)learning steps?
IMO, this solves both of the issues above because the fact that the setting is accessed through the debug console implies that it is not meant to be used by the average user.
To be clear, I am not suggesting to change what happens when the steps field is left blank. I am just suggesting the addition of a new option.
I assume that you meant “longer than 1d”.
1 Like
I mean, if your first four parameters are longer than 1d, you don’t need the learning steps, so I recommend leaving the learning steps field empty in this case.
In the second screenshot I do have a relearning step and FSRS seems to be taking over the good button anyway.
Also, every learning and relearning card I’ve seen has had the exact same interval for the again, hard and OK buttons. While zero new cards have shown this behavior. There is always a big difference between again and hard for my new cards.
Also, no matter how many times I hit good, the interval stays the same even though the difficulty keeps dropping
So it seems to me like something is going on with (re)learning cards and short term scheduling. It’s almost as if the again interval was used for all the buttons except easy.
Is this expected behavior?
Could you share your FSRS parameters? I will check the code.
0.3863, 1.0069, 4.2149, 16.7678, 5.3000, 1.1565, 0.7169, 0.0483, 1.3346, 0.2126, 0.7328, 2.0790, 0.1078, 0.3082, 1.4908, 0.0000, 3.0730
