Lapse Interval - maximum interval

A lot of people are under the impression that cards don’t have to be relearned entirely after a lapse. Many people have no problem with using a lapse new interval of 50 % and more, which significantly lowers workloads. Especially, if otherwise the “baby” learning steps of Anki are used.

Some competitive spaced repetition software, as well as SuperMemo in its first years, experimented with re-learning hypotheses based on ancient wisdom of learning psychology, e.g. by halving intervals after a memory lapse. Current data shows clearly that this is a harmful and time-wasting approach.

This is sensible as well: an atomic, isolated memory, can actually be forgotten (check the difference between stability and retrievability) entirely and thus has to be relearned. There is no way that you will still know that 3 year interval item after 1.5 years (new lapse interval = 50%) if you have forgotten it after that mentioned 3 year interval.

But for short intervals (let’s just assume < 1 year), the new interval parameter actually works pretty good. (As only retrievability was an issue, not stability)

So, why not also include a maximum interval for the new lapse interval to fall back to the usual relearning steps once the interval gets uncannily high?


Hi p4nix,

No objections to the idea itself, but I’m a bit concerned about the number of options we have already - it can overwhelm new users, and some users unfortunately feel like they are missing out if they don’t tweak everything in sight.

One possible solution in the future would be to allow for some customization via Javascript running in a sandbox. It would mean such changes aren’t limited to the desktop, and would allow for more flexibility than a set of pre-defined options.


An idea to not overwhelm new users is having a “Basic settings mode” and and “Advanced settings mode” just like a simple calculator apps do with basic calculator and scientific calculator.

Then newbies could tweak settings on the go without messing around and hindering scheduling data in future states. On the other hand, seasoned users could tweak the settings even further without relying heavily on add-ons.


I can understand the concern about the number of options.
But the main issue I currently see is that the current default algorithm will only be suitable for a limited number of use cases. This is why optimisation of the algorithm is obviously critical. If the optimisation of the algorithm for each user could be done “under the hood” and automatically, we wouldn’t have the problem with complexity.

Being able to customise the scheduler via JS in a sandbox would give the user more freedom, but that’s even more of a nightmare in the end I guess.

I am currently trying to research SM-17. Maybe it is possible to implement it, we will never know without trying. I am aware that a lot of other things are on your todo list, but I believe eventually the issues of the modified SM-2 will have to be tackled. It just doesn’t feel right having to choose between SM and Anki, when Anki clearly is the better choice, if it weren’t for the algorithm (for incremental reading, I have some good and realistic ideas for an add-on).

On the verge of being off-topic: What are your thoughts about enabling the user to share their revlog data via AnkiWeb to enable independent memory researchers to do their thing?


It’s a possibility in the future, but not the highest priority at the moment I’m afraid. Mnemosyne publish a dataset, which could be used in the interim.

I wrote an addon for this if anyone wants this feature: Maximum Interval After Lapse - AnkiWeb.

I tested it, but it’s still new so let me know.


No shared item found with provided id.