We’re basically simulating that in the first 1000 day, there are 20k total cards and you’re studying (adding) 20 new cards every day.
Btw, I feel like there is still a bit of a backlog (20 new cards, 200 review limit), otherwise idk why True Retention would drop later
And how big is the backlog 
Is the backlog arbitrarily set or is it a product of the conditions set for the sim.
It depends on the ratio of new cards per day and review limit per day
Ah so it depends on the conditions arbitrarily set for the sim then I don’t know how feasible this is, but maybe try then the sim for different ratios of new cards and review limit per day then… 
The conclusion will be the same, aside from when review limit is like 9999
1 Like
Can you say what the conclusion is of this rollercoaster of a thread for @L.M.Sherlock and @rich70521
- If the user can do literally any number of reviews per day, the order doesn’t matter.
- If the user can only do a limited number of reviews per day, then Retrievability Descending is the best in terms of spending the least amount of time studying while still maintaining retention at the desired level.
Additionally, I made a table of sort orders that can maintain retention at a constant level even if the user never gets through the backlog.
1 Like
Wow 300 replies in just a few days. That is a lot to process. OP could you please summarize in first post?
See my comment right above yours
Can you update this with PRL as well
The data in tables shows a different picture. I am also looking at seconds per memorised card as I think that’s a better metric to look at.
1 Like
Literally, where did you pull this out from? There was a simulation with no review limit (9999 limit?).
Just in case, you calculated reviews/new correctly right? I mean reviews here equals (review cards + new cards), right? (interday learning too but we won’t have them in the simulations).
I’m not sure. Looking at Jarrett’s code, it’s not clear to me whether first reviews of new cards also count for the same day. I think they do
Seconds per memorized card is important, but if an order cannot maintain retention at the desired level in the presence of a backlog, it’s not very useful
PRL Desc is very close (but maybe noise is more) to R Desc. Maybe should Dae consider adding it
I actually have an idea for what to do. Make the default sort order “Automatic” and then have another called “Custom”. The latter allows you to do all sorts of customisation and the former is decided by Anki.
Here, as you can see R desc and R asc both have terrible knowledge acquisition rate. I think if we do use my idea we can use conditionals that allow us to change the order according to user needs.
This also means people don’t need to understand how to actually use the sort orders.
