I suggest making the default value of max. interval 3650, or roughly 10 years, instead of 100 years. Or maybe even 1825 (5 years). I never understood why the default value is so high. Nobody will be using Anki for 100 years, and the time savings from setting it so high are negligible.
Here’s my reasoning for lowering the max. interval:
Suppose you spend 10 seconds per review. If your max interval is one year, the minimum amount of time that you will spend per year is, well, 10 seconds. If your max interval is 100 years, you will spend less than 10 seconds per year. But do you really care about saving 10 seconds per year? Probably not, and definitely not as much as you care about actually knowing the content of the card.
If your max. interval is 1 year and you immediately forget the card after reviewing it, you will spend at most one year of your life not knowing it. If your max interval is 100 years and you immediately forget the card after reviewing it, you will spend the rest of your life not knowing the card. This is, of course, undesirable.
I’ll make the graphs for different values of max. interval using the FSRS simulator, but it will take a lot of time, so it probably won’t be done today.
Max interval of one year would eventually lead to not reviewing new cards, as all reviews will be taken for reviewing already existing cards. For example, I have more than 100K mature cards (and I deleted stuff across the years), that woul require 100K reviews per year assuming no lapses for no bennefit added in retention, or super marginal.
Increasing intervals is the nature of spaced repetition. Eventually the interval will outlive the user’s lifespan. There is nothing wrong with that. The user’s memory reached a stability long enough that does not require more repetitions. Indeed, small percentage of cards like this will be forgotten as the user is human (true for any stability range).
Just to note that I also use Supermemo, and the intervals get higher a lot quicker. With just 3 years of use, I already have cards that I will probably see again once or twice in my life.
I used the FSRS simulator to simulate 100 years of Anki with a collection that has 30 thousand cards. 100 000 cards just took too long. As for FSRS parameters, I used parameters from one of my decks. Desired retention=0.9.
maximumInterval = 365 # only this setting was changed between simulations
new_cards_limits = 10
review_limits = 9999
max_time_limts = 10000
learn_days = 36500
deck_size = 30_000
As you can see, the amount of time the user would have to spend per day didn’t change much when I decreased max. interval from 36500 (100 years) to 3650 (10 years). But when I switched to 365 (1 year) something strange happened - the workload given by FSRS became greater than that of the old algorithm, and retention exceeded desired retention (not shown here). So it seems that setting max. interval to 365 messed up FSRS. I suggest to make max. interval 3650 (10 years) by default.
EDIT: there was a bug in the simulator, the max. interval setting wasn’t applied to Anki’s algorithm. I will redo this later.
Just to confirm as I see 30k cards for all three grahps, 10y of Anki has 66,650 cards, and 100y years of Anki has 395,000 cards? That is assuming the user never deletes any card and the 10 daily new cards is net and absolute.
A suggestion would be to add the three graphs for different number of new cards. 10 daily new cards is not that realistic for a language learner, specially multilingual.
I don’t understand what you’re trying to say. 30k cards is the deck size. It means the simulator will simulate 30k card histories. No more cards are added.
A suggestion would be to add the three graphs for different number of new cards.
This is already taking >3 hours for one chart, and I’m planning to add a 5 year max. interval chart as well, for a total of 4 charts (max. interval = 100 years, 10 years, 5 years, 1 year). I’m not very enthusiastic about running this for many days.
The criteria followed is that the user introduces 10 new cards each day. The plot you did is just to keep your 30K current deck stable in long-term memory in the next decade or century. The graphs decrease asimtotically, which is to be expected for a deck that is static and will never grow and will stabilize to whatever average daily time is required to review a 1/365th of the collection in year, as all cards will high the cap (assuming no lapses); the same for the 10y cap: daily review time of 1/3650th the collection.
You need to take into account the new cards that will be added while keeping the older -already existing today or not- ones. This is where a 1y interval cap makes a difference and eventually capping the new cards per day to zero as 100% of the reviews will be dedicated to already existing cards.
Anki’s algorithm does not scale well with large number of cards in the collection and in the long-term intervals. To my knowledge FSRS performs better for intervals below a year, in the sense of requiring less reviews for the same retention rate, but when the interval exceeds one or a few years is a different story. At least for now. I know long-term Anki usersusers eventually face this issue. Some of them decide to suspend or delete cards after a certain interval threshold such as the Migaku add-on.
I used the FSRS simulator (as I mentioned previously) to simulate 100 years of Anki with a collection that has 30 thousand cards. Desired retention=0.9.
maximumInterval = 365 # only this setting was changed between simulations
new_cards_limits = 10
review_limits = 9999
max_time_limts = 10000
learn_days = 36500
deck_size = 30_000
However, I think it’s easier to interpret if I just report average total time per card:
Average total time per card is basically “how many minutes you will spend on this card over it’s entire lifetime”.
It’s worth mentioning that for max. intervals of 100 years and 10 years, retention accurately matched desired retention (in FSRS), but for 5 years and 1 year retention exceeded desired retention.
Overall, I think 10 years is a reasonable max. interval.
While this thread is certainly insightful, I dare say it’s pretty much inconsequential if the default is 1 year or 10 years. The small share of users studying with Anki for more than a decade would have tweaked the settings to their liking by then, anyway.
I’m inclined to agree - I don’t really see the point of moving it to 10 years, as it does increase the workload, and I think people who have been using Anki that long have probably got their own ideas about the ideal limit. It also doesn’t solve the initial issue you reported, where users were assuming the wrong unit of measurement instead of reading the help text. The current setting is effectively “off”, and users can enable it if they wish.
