Interface for writing spaced repetition prompts inline while reading web content

The mnemonic medium can be extended to one’s personal notes; this has given structure to my prompts and has helped me write them more fluidly and confidently. But switching back and forth between reading material and the notes still has friction. And I still sometimes feel disoriented when reading prompts in notes because Prompts written in prose notes about source material lack context. My instinct is that some kind of inline interaction would help here.

In 2022, I implemented several working Orbit prototypes of this workflow; see Project - summer 2022 demo (inline marginalia and suggested prompts in the mnemonic medium) and Project - spring 2022 demo - a peritextual mnemonic medium. I also shipped a working version of this workflow for Project - collaboration with Delta Academy.

On 2021-12-22 I prototyped a workflow like this:

• I write inline prompts using Hypothes.is (using the same format as described in My implementation of a personal mnemonic medium)
• An Obsidian plugin imports those into my Markdown notes
• Orbit’s note-sync imports those prompts

Log

2021-12-30: Ozzie Kirkby tried to make machine-generated prompts match mine! His notes.

• A thoughtful question from Ozzie helped me notice: “For the most part, I wrote prompts in realtime as I read. I noticed that I felt less resistance to doing this than I usually do while reading. I think lowering the friction made one-pass reading more tractable.”

2021-12-27:
Just capturing some observations so far:

• I’m surprised by how much better it feels to write these prompts inline, compared to writing them in my notes.
  • I notice that the context switch feels lighter-weight. Switching to my notes and back requires orienting myself in each environment: where was I? Where does the next prompt go? And then when switching back: where was I on the page?
  • The highlight affordances give me some very rough impression of “coverage”. When I was writing prompts in my prose notes, I think I was doing some mental bookkeeping to keep track of what I’d written a prompt about and what I hadn’t.
  • On a few occasions, I needed to “find” the prompt I’d written about some concept, and I could easily do that by scrolling to the place in the text where the concept was introduced, then click the appropriate highlight.
  • When I’m looking at a prompt I’d already written, the Hypothes.is interface is such that I’m almost always looking at it in context—I see the “nearby” text at the same time. This is great!
• In principal, a huge advantage here should be spatiality—the physical association between prompt location and text location. Unfortunately, this effect is diminished in my current prototype because Hypothes.is annotations aren’t really spatialized, like Google Docs comments: they sit in a linear list in the sidebar.
  • This means that I can’t simply scroll to a part of the page I’m interested in and see the “nearby” prompts. Instead, I have to scroll to a part of the page, then click a random highlight to scroll the sidebar to the right “neighborhood”, then browse around from there, with no clear delineation of the “edge” of the neighborhood.
  • I’m surprised that they don’t distribute the comments spatially, given that they went to the effort of building very elaborate “pointers” in the gutter of the sidebar which are indeed spatialized.
  • Those pointers don’t solve my problem, unfortunately: they’re too clicky, too fiddly. And their interaction suffers from The detail-in-context problem.
• The editing interface makes it clunky to add several prompts about the same passage. This doesn’t bother me that much yet.
  • It should be possible to write several prompts in a single H comment, but the formatting of the newlines separating the prompts is lost at some point in my convoluted import pipeline. Probably I can fix this with some effort.
• I don’t like that Readwise only imports from Hypothes.is only once per day. That makes it difficult to fluidly move between writing inline prompts and writing open-ended prose notes, since the note for the web page in question likely doesn’t exist yet.
  • Small friction: prompts for the various chapters of a web book will be split into different notes, whereas I’d probably put them in the same note if I’d written them by hand.
  • There’s an Obsidian plugin which imports directly from Hypothes.is; I’ll try to switch to that.
  • Also, Readwise doesn’t seem to import Hypothes.is’s document location metadata, so I can’t easily jump back to a page location from a note excerpt.
  • … OK, done—switched to importing highlights directly from Obsidian. Now latency is low, and I’ve cut one link in the chain out of the loop. Also, it appears that I can now successfully write multiple prompts in a single H highlight! Excellent.
• Obsidian still has enough rough edges that I find myself not wanting to use it to actually write… but I’m OK with using it just to run the import plugin. Probably someone has written some command-line script I could just run instead?