I’d like to propose a feature idea for Anki — a “Memory Strength Score” derived from FSRS parameters after optimization.
As we know, FSRS uses 21 parameters that are adjusted during optimization to fit a user’s review history. These values indirectly reflect a user’s memory performance. But currently, there’s no clear way to interpret whether the updated parameters indicate improved cognitive performance — for example, stronger memory or more efficient recall patterns — especially when desired retention (DR) is held constant (e.g., DR = 90%).
I propose that Anki (or FSRS-based tools) implement a metric that compares two sets of optimized FSRS parameters — Parameters A (from earlier data) vs Parameters B (from a later optimization run), assuming the same DR — and outputs a score (0–100) representing whether the user’s memory performance has improved or weakened.
This would provide:
Motivational feedback for high-discipline learners (e.g. users doing 1000+ reviews/day).
A way to quantify neuroplasticity and learning progress, especially during long-term study regimens.
Insight into whether a learner is retaining more with less review load over time.
Well-known polyglots and autodidacts (like Kato Lomb) often report that memorization becomes easier over time. This feature could offer quantifiable confirmation of that trend based on data.
My question is:
Is it theoretically possible to design such a score — one that analyzes FSRS parameters and compares their impact on recall intervals, difficulty decay, or workload — to represent “memory strength” under a fixed DR?
I’m not pinging FSRS developers directly, but I’d love feedback from anyone technically familiar with the scheduler or interested in building such a plugin or model. Thanks for reading.
TL;DR: I’m proposing a “Memory Strength Score” feature for FSRS/Anki that compares your FSRS parameters over time (under fixed difficulty rating like 90%) to show whether your memory has improved. This would give users motivation, show cognitive progress, and reflect neuroplasticity from long-term discipline. Is it theoretically possible to build such a score from FSRS parameters?
Only issue is that even though parameters are calculated from your earlier reviews, they do not fit perfectly well with them. And the more reviews you have the better the params fit. Also, parameters have nothing to do with DR, you don’t need an adjustment for it.
Now, I did propose similar ideas before like yours. There’s a knowledge acquisition rate that you can calculate and FSRS Helper add-on includes that already. I later suggested we can adjust this value for DR but nothing happened after that.
One other feature that we can use is the simulator, you simulate with old/new params and check the memorised/review count for each simulation. This is perhaps the best way.
Parameters don’t necessarily reflect how good your memory is intrinsically because they also depend on how easy the material is for you.
But more importantly, I recently asked Jarrett to add a certain similar feature to the FSRS Helper add-on. Sadly, due to technical limitations of Anki add-ons, it cannot be done.
My idea was to compare the rate at which memory stability increases (at a fixed 90% desired retention) for different users and calculate the geometric mean of all increases. If the geometric mean is, for example, 1.5, that means that memory stability increases by 1.5 times after a review, on average. So if your geometric mean of the increase in S is 1.5 and someone else’s is 1.8, that means their memories become stable faster than yours, though, again, it is confounded by the difficulty of the material. Then we can do this for all 10 thousand users from our dataset and calculate percentiles. Then we can compare the geometric mean growth of S of a given user to known percentiles to display something like “Your memory matures faster than that of 77% of users!".
But yeah, he can’t implement that in the add-on.
Does that mean that even if I change my DR, it doesn’t affect my FSRS parameters?
Is it possible to use some other data or metric from FSRS/Anki for this purpose? Or is it too complicated to design the parameters themselves so they’d be useful for direct A-to-B comparison?
Expertium, I appreciate your response!
But if the material stays the same, and I do around ~1000+ reviews over a month, why wouldn’t it work to compare my FSRS parameters with how they were a month ago? Even if not the parameters themselves, is there maybe something else in FSRS that could be compared instead?
Maybe this could be implemented directly in Anki at some point, at least in the long run? At the very least, it seems there’s room to improve FSRS by adding new features.
Could you please explain whether it’s actually possible to compare a user’s memory performance over time with current Anki/FSRS? In other words, is there any way to track if my memory is improving, staying the same, or even getting worse?
Comparing with other users is great, but comparing with yourself would be more precise. Is that possible? I don’t fully understand how it works, so I’d really appreciate it if you could explain the details.
As an ordinary user, it seems to me that FSRS has different parameters that reflect how my memory works. So why can’t I just record them in a Google Sheet and track them over time? I’d really like to find out if there are any changes in my brain (signs of neuroplasticity) based on how it responds during periods of intense study versus periods with little or no studying.
The approach that I suggested, with the geometric mean of growth of memory stability, probably works for the same user too, though I’m not 100% sure about that.
But also, I’m not convinced that this is worth implementing in Anki. I can think of a few useful FSRS-related things that may be implemented in the future, and this isn’t one of them.
I’m not trying to compare my performance. I want to track whether my own memory is improving over time, specifically, whether my ability to memorize and do Anki reviews is getting better or worse.
For example, there’s an add-on that shows retention rates for each learning step.
I’m trying to use this add-on to see if my brain has adapted to a certain learning step and whether my retention has improved over time.
But I don’t know how to apply this to long-term retention, especially when using FSRS.
Performance seems like an adequate and appropriate way to measure that.
You’re supposing that using Anki could improve your ability to memorize – like that’s a muscle that can be strengthened. I’m not sure how well that’s borne out by the science. [It certainly seems like a popular idea in the field “brain training” apps, but I should leave that to be addressed by those with more expertise in that area.]
But even if it is, does that seem measurable? Is it that your memory has improved? Or have you gotten more effective at note/card creation, or have you become more familiar with the overall material you’re studying, or have you refined your study process to make it more efficient, etc.? There are too many variables you’d need to eliminate.
Navigation-related structural change in the hippocampi of taxi drivers.
London cab drivers, who memorize complex maps of the city, had enlarged posterior hippocampi (the brain memory center). This was not genetic, they developed it through training.
Meditation experience is associated with increased cortical thickness.
Meditation studies show increased gray matter density in the hippocampus and prefrontal cortex after sustained mindfulness training. These are regions directly tied to attention and memory.
Kato Lomb, a hyperpolyglot who taught herself over 10 languages, repeatedly emphasized that memorization got easier the more languages she learned.
She described her memory as improving with each new language.
There are stories about the Mauritanian memorization method, which involves extreme rote memorization. People who use this method report that their memory improves over time. However, this is not well-researched.
My theory, which I am currently testing, is: what if it’s possible to improve memory? If that’s true, then I should be able to set a relatively large single learning step above 10m/15m and track my % retention rate for that learning step using an add-on.
I’ve chosen 20m as my learning step. My initial retention is around 50%. If my assumption is correct, then after 30-100 days of daily and perfectly intense study, my retention should increase. For now, I am at the initial stage of testing.