<h2>On self-experimentation in spaced repetition</h2>
I <a href="on-using-anki.html">love Anki</a> but am sufficiently devoted to spaced repetition to want more control over my training. I now spend over half my memory-training time in a custom tool.
This tool uses a much simpler scheduling algorithm: in the case of a correct answer after an interval of length L, the question comes up for review e * L later, where e is Euler's constant. So, if I answer a prompt correctly 5 days after I last saw it, it will come due in roughly 13.6 days. (Except if L is very short--there's a minimum interval of one day.)
This is a very primitive system, and I'm not interested in arguing that it's better than any existing algorithm. It has these virtues, though:
<ol>
<li>It's simple enough to have implemented and tested fairly quickly. Minor and even moderate flaws in the system are acceptable, but major flaws are not.</li>
<li>It's working surprisingly well. The reviews it schedules tend to be tougher than the ones Anki suggests on a given day.</li>
<li>Its flaws should be easier to fix in a natural way. I have often overlooked this feature of simpler systems. Ockham's razor is often presented as a guide to truth; other exhortations to simplicity are often justified in terms of cost. This is more a benefit of natural diagnosis.</li>
<li>It provides a benchmark against which more complicated algorithms can be tested. (This is a useful feature I associate with <a href="http://tangotiger.net/marcel/">Marcel</a>. If any baseball analysts are out there looking for a project, I think there's some relatively low-hanging fruit available in optimizing the various constants that appear in the Marcel system.)</li>
</ol>
I suspect that the best systems are likely to use machine learning in some form. It took me a long time to come to believe this: existing algorithms have their roots in, for example, systems for <a href="https://en.wikipedia.org/wiki/Leitner_system">moving physical cards between boxes</a>. These algorithms make a certain kind of sense. Their features are each intuitively motivated, and there are enough of them to make you think that the real-world phenomenon is being captured sufficiently well.
But we have computers now. The way to work with the computer, rather than against it, is often to keep it very simple until it can be totally inscrutable.

On self-experimentation in spaced repetition

<h2>On using Anki, again</h2>
Thinking in terms of dependencies is systematically underrated not just in software development but in using personal-management software. Consider this (stylized but fairly common) exchange:
A: "Where should I write arbitrary documents?"
B: "Use TeX! Liberate yourself! Produce arbitrarily great documents!"
A: "Sounds good! How do I change the font?"
B: "Ahahahahaha no."
There are constraints you determine and choices you make, and some the software makes. These can interact in complicated ways. You set up MyFitnessPal, and then it gives you a budget. An email system forces you to think in terms of folders / VIPs / inboxes / whatever, but you determine when and what you email, whether to auto-archive, which clients use which accounts, and so on.
This isn't just about ergonomics: it's about what causes what in your life. This matters a lot with Anki. An anonymous reader asked me how exactly I used Anki, and in particular whether it makes sense to commit to using X minutes per day in Anki.
I'd hate to do that, because I need Anki to sit downstream of my cognitive goal-setting. An underrated benefit of using Anki is that it tells me how much of my mental resources are free and available. But I want it to tell me what my capacity is, not how to use it. It's my job, not Anki's, to figure out whether to use slack immediately, or carry it forward and have some room in my memorization budget the next time some valuable fact set comes along.
<a href="on-using-anki.html">Again</a>, the best argument for adopting a spaced-repetition regime is not that there exists an exotic, beneficial system of mental training--as much as one might be interested by such a system. It's that we all already want to remember things, and spaced repetition is the most efficient way to do that.
For me, Anki is a good driller but a bad personal trainer; DEVONthink is a great "personal datalake" but a bad life-hierarchy-describer; Goodreads is useful at ingesting data and not useful at describing that data. I've failed with these (and with much other software!) when I've let them determine something they shouldn't or gotten in the way of letting them determine something they should.
Each of those failures was, fundamentally, a failure to properly manage the dependency graph for my mind correctly.

On using Anki, again

<h2>On using Anki</h2>
Anki is software that helps you remember things. It is used largely by people with idiosyncratic and extreme self-development goals. Much of what you can read about Anki therefore recommends idiosyncratic and extreme behavior.
But we all want to remember at least some facts: that thing we keep needing to look up at work, our cousins' birthdays, or whatever. And Anki is the most efficient way to remember facts. So Anki is useful for just about anyone.
That argument is the memory-management analogue of "if you can't afford a good lawyer, you certainly can't afford a bad one." If you don't want to spend effort memorizing things, software that minimizes the effort per thing remembered is what you want, even if other people use the software to increase the denominator when what you want is to decrease the numerator.
<h4>Miscellaneous notes</h4>
<ol>
<li>Don't worry about duplicating cards. You'll simply see each of the duplicates less often, and this will approximate there being just one card with that information. It's probably better to have cards duplicating information but presented slightly differently, so that your brain doesn't latch on to the presentation so easily.</li>
<li>All else equal, switch up the physical context in which you study cards (again, to encourage your brain to learn a connection between a prompt and a fact, not a connection between some wallpaper and a fact).</li>
<li>I like using fewer decks, more loosely organized, so that (i) I don't spend as much time clicking between decks and (ii) there are fewer accidental patterns in the clustering of cards I review.</li>
<li>When I'm reviewing a new batch of cards, I "jitter" the review times--e.g., by finding some to mark as "easy" and others not. This makes them pop back up on different days. (Adding some random variation to the review interval seems exactly like a feature Anki would already have, but I've scoured the preferences and am pretty sure it's not there.) [Update 2021-12-19: There is such a feature, and it's called "fuzz factor," and <a href="https://faqs.ankiweb.net/the-2021-scheduler.html">its behavior is changing</a>. I still find myself needing to jitter things manually. Thanks to David Allison for the information.]</li>
<li>Unless you are in med school or a language learner, most publicly available decks are not as good as what you can make yourself. It's probably most time-efficient to simply roll your own. An advanced clipboard manager is your friend here (and, for that matter, everywhere).</li>
<li>Don't put too much information in the prompt: again, it's an invitation to form an association with an incidental pattern. This is a particular problem with automatically generated decks.</li>
</ol>