I spot checked some ~100% cards, and there are cards there where I have only ever pressed Good or Easy, so I really don’t think their difficulty is accurately represented.
I have never misused Hard. I used to be really bad at pressing Easy, but I have gotten better at this lately. The deck is over 10 years old, and I’ve had year long streaks of both consistent reviews and no reviews at all. I did have New Interval set to 50%, which I know now was a mistake.
From reading the formula it seems like not using Easy enough is what caused this. It does however still not make sense for the aforementioned cards to be so high, and it feels like I’m going to be doing a lot of junk reviews because of this.
Am I correct in my analysis, and is there anything I can do to remedy this (from reading a few other threads it seems like the answer is going to be no, but I figured it’s at least worth asking)?
How long has it been since you re-optimized your FSRS parameters?
FSRS D[ifficulty] doesn’t necessarily always match up with our innate sense of what’s hard and what’s easy for us. [Perhaps this article will explain it better for you: A technical explanation of FSRS | Expertium’s Blog .] It’s just one of the values that FSRS uses to schedule your cards accurately. So a graph like yours doesn’t necessarily indicate there’s a problem.
I think the first time was a few weeks ago, and the second time was yesterday. I also slightly overestimated the time I’ve been using FSRS, it seems to be exactly two months.
I don’t expect it to match up perfectly, but in my case D could almost be a constant.
That was way worse than the one I linked!
To me S barely increasing for most of my cards definitely feels like a problem, but maybe I’m overestimating the impact of this. I get that it’s probably not that useful to obsess over FSRS values, but I also can’t shake the feeling that I’m stuck in the ease hell that FSRS promised to save me from (even though this version of it might be preferable to the SM2 version(.
Also, it turns out this was probably not true. I can’t find that card, and given the fact that I barely used to press Easy I probably just misread the review log as zero-indexed. There are definitely cards that I have never lapsed that have ~100% difficulty, but this is consistent with how I would expect it to work from the math.
Was yesterday’s optimization before or after your screenshot?
And as long as we’re completing the picture – what are your parameters now, as text, please?
I understand it feels that way – because bar graphs do that. But even just eyeballing it, you’ve got around 5K cards out of ~20K cards that aren’t in that 95%+ bar. That’s not nothing. And D is just one piece that goes into scheduling your cards.
While reasonable minds can differ on whether “ease hell” ever even existed – the fundamental difference from SM-2 is that FSRS grows and changes. Each month when you re-optimize, FSRS reconsiders what it knows about your memory curve and updates the memory state of your cards.
I wonder – How has your retention been for the past couple months [Stats > Retention]? Has it been up or down from before FSRS? Has it been meeting your Desired Retention? I also wonder if you’ve “rescheduled” your cards since you enabled FSRS (I’m not recommending that, but it helps to know if it has happened).
That’s true, and I’ll be the first to admit I’m probably more concerned about this than I should be, but I’d still like to understand it better even if it turns out there’s nothing I can/should do about it.
My DR is at 80%, and last week I was at 77.4%, so it seems to be getting close. I spent the second half of last year catching up with a huge backlog, so I think that messed up the stats quite a bit, but according to the stats screen my retention for the last year was 69.3% (all time is 76.3%). Is there a way to check what I had specifically before FSRS?
Yeah, things like this are part of the reason I use binary feedback: you either pass (good) or fail (again), no ‘hard’ or ‘easy’. It works very well for me, and it greatly reduces decission fatigue. See the second question of the FSRS FAQ:
Q2: I only use Again and Good, will FSRS work fine?
A2: Yes. In some cases, FSRS may even be more accurate if you only use Again and Good.
When you’re working through a backlog, your retention can be a bit lower, but FSRS doesn’t get confused by that, so I wouldn’t. It knows how long it’s been since the last time you studied the cards, and it’s making allowances for that.
There isn’t really. [1] Usually it’s enough to get a general sense if you compare a longer time-period (a year or all history) with a shorter time-period where you know you’ve been using FSRS.
Unless perhaps one of the “time machine” graphs in the Search Stats Extended add-on has that? ↩︎
pass (good) or 
fail (again), no ‘hard’ or ‘easy’. It works very well for me, and it greatly reduces