Intervals do not always match desired retention

Hello

It has been brought to my attention that for some reason, my intervals do not match my Target R%, and that I am getting intervals too far ahead.

At the moment I have to use filtered decks, but sometimes this causes problems too.

I have a deck of 1156 cards, with 120 reviews due today. My desired retention is 90%, my FSRS parameters are:
0.1236, 0.2864, 0.5138, 0.8452, 7.1268, 0.3551, 2.3149, 0.0033, 1.1783, 0.4186, 0.7260, 1.8196, 0.1104, 0.2487, 2.2342, 0.2319, 3.0685, 0.8349, 0.0926, 0.1402, 0.1026
I am on Version ⁨25.05 (ad073ab1)⁩.

Ι have optimized my parameters, with log loss of .3691 & RMSE bins of 2.60%. I cannot optimize further. I have some difficult cards that once I get wrong, I get a 1 day interval. If I get it correct the next day, it’s a 2 day interval next. However, this 2 day interval is too long, because the retrievability drops far below my desired retention, the target R% as well. I am using again and good, do not have hard misuse history.

When I use a filtered deck with the option prop:r<0.9, the reviews total up to 139. 56 from the normal deck & 83 in the reviews.

For some cards, (in the column after the stability, which is the hours, that is Target R%)
For some reason, FSRS wants to override showing it to me at 90% and instead of giving me an interval of 2 days. For the cards with 1 day interval there, those are because I had just gotten them wrong. The cards with 2 day intervals, were done the day before. It does not make sense to me why when I hit good, I get a 2 day interval, even though it would cause my retrievability of that card to drop far below 90%, to 40% even!


For some reason it is coded that no matter what, the good interval must be 1 day more than the hard interval. This is overriding FSRS, what is the point then?

1 Like

I am not trying to say there isn’t any problem worth discussing, but it’s important to mention that the target R column is added by the FSRS helper add-on and it doesn’t support FSRS 6 yet. So, the value of target R would be incorrect.

1 Like

Good point, I did not think of that.

Here is more that I will add to the thread as well.


this card gave me a 2 day interval. I manually rescheduled it to today. Today, it is at 87% retrievability, so how can FSRS expect me to be fine with 2 days?

@L.M.Sherlock is it true that the target R column is not based on FSRS-6?

It is based on a new API which hasn’t been released in the latest beta version.

1 Like

Now I have made an interesting observation.

The “normal” (unchanged) due date makes retrievability of 1 card at 98%.
If I set the due date to 1 week before due date (today), the retrievability becomes 87%

If I set the due date to 1 week after due date, the retrievability becomes 100%

- original
- 1 week before
image - 1 week after

how did you set due date? 7 or 7!?

Edit: OK, I figure it out. The retrievability column is calculated based on the elapsed days and stability. The elapsed days are equal to Today - (Due - Interval).

If you didn’t append ! after the interval, the interval is not changed. So the elapsed days are wrong.

However, even if you appended ! after the interval, the interval is still incorrect. For example, if today is 2025-05-26, the Due is 2025-05-30, the original interval is 7, and you set 1!, then the new interval should be 3 days instead of 1 day.

@dae should we fix it?

This is an old known issue.

2 Likes

OK, I see the closed PR. I will try to re-implement it.

2 Likes