Improve hard and easy intervals

[AnKingMed]

As discussed here based on @Expertium's research, currently FSRS is better with just 2 buttons than 4.

As I've been using FSRS more I've regularly felt that (at least according to my patterns), Hard was too far out and Easy was also too far out. This is of course subjective.

Hard I've noticed is often very close to Good. 1.3 vs 1.5m for example. In that case Anki would have done 21 vs 45 days. I imagine many people's button habits are at least roughly similar to how Anki spaced things. I could be wrong.

Easy also feels much too far out, especially on the first interval. Sometimes it's like 15+ days on the first interval. I generally want to see things at least this week if I'm studying it for the first time. Of course, what I want and what actually leads to the retention I desire are two separate things, but I'm just trying to offer an observation in hopes of improving things

My collection if it's helpful: https://drive.google.com/file/d/1Ibvih-Ps-XDMFTrK2zHu6YTMhYFFTeU6/view?usp=drive_link

[user1823]

Easy also feels much too far out, especially on the first interval. Sometimes it's like 15+ days on the first interval.

It may be possible that you "need" a smaller S0 for Easy. That can be found out using the collection data, once L-M-Sherlock analyses it. But, I want you to read Q5 of the FAQs.

I also wanted to share with you my personal experience. The first interval for my deck with the Good button is 8-12 days (depending on fuzz) even with a desired retention of 0.94. At first, I also thought that it was too long. But over time, I have come to realise that this interval is not that long and it saves me a lot of unnecessary reviews that I would have done if the interval were shorter.

[user1823]

That being said, FSRS is still not very efficient in dealing with Hard & Easy buttons, especially for review cards.

The current Hard & Easy factors are just makeshift solutions. Properly dealing with these buttons will require more elaborate factors.

[AnKingMed]

For sure. I admit it's possible that the longer intervals are potentially actually good, but given that 2 buttons is greater than 4, I wanted to try and offer some observations in hopes that we could improve that model. It's very possible my observations are very off. I feel like I use hard and easy consistently the same way though so there should be somewhat of a pattern. The challenge of course is then extrapolating that to thousands of users' habits

[Expertium]

I have an idea how to make FSRS more flexible, but I'm currently testing other ideas too, so I'll probably start working on Hard/Easy next week.

[Expertium]

This isn't the idea that I mentioned in my previous comment, this is another idea. I wonder what you all think
#461 (comment)

[Expertium]

As for the idea I had in mind, I benchmarked it and it didn't work.
#461 (comment)