How Podsage decides when to bring a note back

Forgetting has a shape

In the 1880s the psychologist Hermann Ebbinghaus spent years learning lists of nonsense syllables and testing how much of them he kept. We forget quickly at first and then more slowly. When a researcher repeated his experiment in 2015, the curve came out the same shape.¹

Writing something down does not stop that curve. If anything it steepens it, because once a note exists you can stop thinking about it.

Spacing is what works

A review of more than 300 experiments found that spreading study over time beats cramming it, and that the best gap grows with how long you want to remember.² Looking things up again is not enough on its own, either. In a well-known study, students who were tested on a passage remembered more a week later than students who simply reread it.³ In a later review of ten popular study techniques, spaced practice and self-testing were among the few rated highly useful. Rereading and highlighting were not.⁴

Spaced repetition turns that into a rule: look at something again just as it starts to fade, then wait longer each time it holds. Here is exactly how Podsage does it.

1. Each note carries a small memory model

A note only joins the schedule once it is enrolled. Podsage does not review everything you write by default. An enrolled note carries four numbers:

2. Reviewing updates the model

When a note comes back, you answer in one of three ways, and the numbers move:

• Forgot. Difficulty goes up 0.2, ease goes down 0.15, and stability resets to 1 day. The note is due tomorrow.
• Partial. Difficulty goes up 0.05, ease goes down 0.05, and stability is multiplied by 1.3.
• Remembered. Difficulty goes down 0.05, ease goes up 0.1, and stability becomes:

stability × ease × (1.2 − 0.4 × difficulty)

That last line is where Podsage borrows from newer schedulers such as FSRS, rather than leaning on ease alone as the classic SM-2 algorithm (1987) does. It means a hard note grows more slowly than an easy one, even when you get both right. At difficulty 0 the multiplier is 1.2. At difficulty 1 it is 0.8, so the easy note’s interval grows one and a half times as fast. Ease is floored at 1.3, so a note you keep failing cannot collapse to never growing.

The new interval is the note’s stability in days, scaled by your “remind me more or less often” setting: ×0.6 for more often, ×1 by default, ×1.6 for less often.

3. The ranking is the clever part

Being due only makes a note a candidate. The daily queue then scores every candidate:

forgetting score = days since viewed × difficulty × importance ÷ stability

Higher means review sooner: a note you have not seen for a while, that is hard, important and fragile, rises to the top. Two of those inputs do not come from your answers at all.

• Difficulty starts from the content itself. A note begins at 0.3 and gains 0.2 for each of these: it is long (over 1,500 characters), it contains code, it contains equations. A dense note is treated as harder before you have ever reviewed it.
• Importance is inferred from your vault and your habits. A note starts at 0.5 and earns more for being pinned, favourited, or filed under Work or Study, for how many notes link to it and how many it links out to, for its reading length, and for how often you have searched your way to it. A note the rest of your vault depends on is reviewed before an orphan.

4. Two deliberate touches

• Neglected notes get rescued. A note with no links at all that has not been seen in 30 days is pulled into the queue even when it is not due, with a priority bonus. Otherwise isolated notes quietly vanish for good.
• Sessions follow the graph. If you are reviewing from a particular note, anything directly linked to it gets 1.6× priority, so a session moves through related ideas instead of jumping about.

The queue is sorted by priority and capped at ten notes a day, which keeps a day’s review a manageable size.

5. And you can always override it

Set a custom reminder, with a date, a time and how often it repeats, and that note skips all of the above. The smart schedule is the default, not a cage.

Arithmetic, on your phone

None of this is machine learning. It is arithmetic you could do on paper, which makes it predictable, and it runs on your phone like everything else in Podsage. There is no account and no server, and the app makes no network requests.

Sources

1. Murre, J. M. J. & Dros, J. (2015). Replication and analysis of Ebbinghaus’ forgetting curve. PLOS ONE, 10(7).
2. Cepeda, N. J., et al. (2006). Distributed practice in verbal recall tasks: a review and quantitative synthesis. Psychological Bulletin, 132(3).
3. Roediger, H. L. & Karpicke, J. D. (2006). Test-enhanced learning: taking memory tests improves long-term retention. Psychological Science, 17(3).
4. Dunlosky, J., et al. (2013). Improving students’ learning with effective learning techniques. Psychological Science in the Public Interest, 14(1).

All posts