Andy Matuschak
patreon
Five years of evergreen notes
🕑 Added 2024-08-01 00:20:15 +0000 UTCLast month, I published an essay that I'd started writing five years earlier. When I began it in 2019, I knew I didn't understand the topic nearly well enough to produce the work I wanted. I still wanted to make progress, though—to articulate and refine what little I could, as a stepping stone.
So, largely inspired by Niklas Luhmann’s Zettelkasten, I started experimenting with note-writing methods. I don’t care about notes for the sake of notes: my aim is a powerful context for thinking. I wanted to create an environment where I could gather my ideas over time, where those scraps could accumulate and evolve with my thinking on the topic, until I could synthesize it into some meaningful whole.
The usual approach to notes doesn’t work very well for this kind of accretion. Notes are typically an ever-unfurling scroll of scratchwork, with new material appended to the end each day. Promising nuggets of insight are too often buried by the next day’s stream of consciousness, rather than being iteratively refined into crisp ideas.
In my approach, I still keep a daily notebook with lots of scratchwork. But when I stumble on something juicy, I try to write a sharp note about that one claim, question, or idea. (You can read some examples here.) The aim is to make each note clear enough to stand on its own, but I fill them with links to other small notes, so I can move around quickly and see how my ideas relate. Each note is aspirationally an “evergreen” entry on its topic, so that I can make progress by accumulating and improving the notes over time, individually and as a network.
Five years later, my recent essay represents a kind of success for this approach—but it also demonstrates some important failures. Now seems like a good time to reflect on my experiments in note-writing environments.
Divergence and maintenance in notes
Why “evergreen” notes? The (aspirational) idea is that each note in your library reflects your current best thinking about that issue. So, as you develop your ideas on a topic over months or years, you can draw upon, and extend, strong foundations. And because the notes are densely linked, when you improve one note, there’s a sense in which you improve all the nearby notes in the network.
The downsides of all this is maintenance burden and a kind of “conceptual debt”. It reminds me of a trope in software engineering. Novice engineers often get excited about how many lines of code they’ve managed to write, how big their projects have gotten. But experienced engineers often view code as more liability than asset. They’d rather have less of it, because they know that every line contains more moving parts to maintain, more pieces which can break. Over time, one part of the system typically evolves in a way which makes another part inconsistent or outdated—creating “technical debt”.
This May, I felt I had finally accumulated enough understanding to write the essay I wanted to write. My work on ”How might we learn?” had given me some new perspective on old problems, and I had a few new ideas ready to articulate. But I didn’t write evergreen notes on any of that new material. I felt ready to dive straight into the essay; writing notes would be an unnecessary detour. So I developed that new material directly in the manuscript.
Having now written that essay, I still have many dozens of notes on this topic. Many of them no longer reflect my most recent thinking: as usual, my views shifted and sharpened while I wrote the essay. Some of those old notes actually seem wrong to me now. And perhaps a third of the ideas in the essay are new, not captured in notes at all.
So, my notes on this topic now seem like more liability than asset. I could update them, of course, but I have no appetite for that right now. It would feel like bureaucratic busywork. The old notes have an air of staleness to them which makes me unenthusiastic about revisiting them. At the same time, I don’t like the idea of just throwing them out: I’m sure I’ll return to this topic someday, and there’s plenty of material in the notes which didn’t make it into the essay.
In a sense, the system served its purpose—helping me prepare, over years, to write this essay. But now my note library has an unmaintained corner which subtly pollutes the experience of using it.
What to do about this? One approach would be immediate visual indications of staleness, like yellowed old paper. It’s tempting to imagine that an AI system could help me automatically maintain these old notes. Unfortunately, that AI would need information that’s hidden inside my head. For instance, if I no longer find an old note convincing, I won’t necessarily denounce it anywhere visible to an AI: I may simply not mention it in recent work. But not all unmentioned ideas are bad—many just didn’t fit into the themes I was discussing.
Notes as scaffolding
Developing new ideas is hard. There are often too many moving pieces to juggle, too many interlocking problems to solve simultaneously. The miracle of writing is that it reduces the number of threads you have to hold in your head simultaneously, because the paper can do some of that work.
Unfortunately, writing is also hard. Note-writing acts as a kind of scaffolding. The environment says: you don’t need to write a ten-thousand word essay. You don’t need to make everything fit into a clear narrative. You don’t need to decide yet what’s important and what’s not. Just focus on that one idea, and express it as clearly as you can. Progress is much more tractable when it comes in freestanding chunks of a few hundred words. Scaffolding like this is precious when tackling tasks beyond your abilities.
Shortly after I started my note-writing practice, I began publishing monthly essays for my patrons. This created another context for writing and thinking, with higher stakes and higher fidelity than my working notes, but still much less demanding than my more serious major articles. Over the past five years, I’ve published well over 150,000 words in these monthly patron essays. And through all those words, I’ve become a much more confident, capable writer. When I started, I needed scaffolding to develop ideas beyond a limited scope. Now I routinely develop new ideas directly in the context of an essay I’m writing. Preparatory notes often feel less necessary. And if I can develop the ideas directly in an essay, I certainly want to do so: the work will generally be more lively, immediate, true.
Like all creative tools, my note-writing practice imposes both costs and benefits. As I become a more confident writer, its marginal benefits do seem to have declined. And as I accumulate more notes—more to link and maintain—I do feel the cost rising somewhat. Perhaps I’ll look back on this note-writing system as a “training” exoskeleton of sorts, one which I’ll someday have fully outgrown. I’ll still be grateful to have had it as a stepping stone.
Notes as public artifacts
In early 2020, basically on a whim, I started sharing my working notes publicly. My original motivation was that in Twitter and chat conversations, I’d often find myself saying “oh, I’ve written a note about that issue”. I wanted to be able to share pointers to individual notes as a way to quickly enrich discussion.
Much to my surprise, this practice ended up creating many more new discussions. People wander my notes, stumble onto something they find interesting, then write to me (or their friends) about it. That network effect has created a lot of new connections and opportunities for me.
One problem with my note-writing practice is that it often lacks a strong emotional context. When I’m squirreling away material for some abstract unknown future, rather than responding to the needs of some specific live project, the experience can feel dull and lifeless. Worse: it can feel like a duty—something I “should” do, rather than something I’m excited to do. But if I know that others will read even my working notes, before they’re subsumed into some more coherent project, that can create a meaningful creative context. That context becomes even stronger when I have particular people or conversations in mind.
The social context can also create some short-term rewards, which I certainly appreciate when my projects stretch on for years. It’s nice to receive warm emails about my notes. In one astounding instance, a college student introduced herself at an event and told me that she’s in a group chat of students at her university that started to discuss my ongoing work, including my notes.
All that said, a socially-motivated context—especially a totally public one—isn’t ideal for rough working notes. If I’m grappling with a fragile idea, that’s already hard enough. I don’t want to be simultaneously modeling and second-guessing what my hypothetical reader will think. That’s its own creative challenge, one better addressed when I have a stronger handle on the material myself. And I don’t want to reward myself too much for working notes. I want to orient myself around producing major essays and projects; the notes are just stepping stones to get there. So, practically speaking, when writing my working notes, I don’t draw much on the social-emotional connection. I pretend the notes are just for me, unless the idea is something I already understand pretty well.
The public availability of my notes makes staleness a bigger problem. Over the years, lots of people read my preparatory notes for last month’s essay and wrote to me about them. Those notes are still out there. Probably lots more people will still read them, even though many of the notes no longer reflect my views. This feels like a double bind: if I don’t update those old notes, they’ll subtly pollute the information environment, and people won’t get the benefit of my best work; if I do update them, I’m signing myself up for chores. One simple half-solution would be to stick a notice banner across all the relevant notes. (Ironically, my notes about note-writing need such a banner.) Unsatisfying.
Notes and spaced repetition
This unorthodox note-writing practice lives alongside my unorthodox spaced repetition practice. It’s been interesting to observe how these two practices have coevolved. Each practice creates a powerful context for honing my understanding, and each aims to combat the tides of forgetting as I work over years. Each practice also has an associated medium—the evergreen note and the spaced repetition prompt—but those two mediums have very different grains.
Evergreen notes are prose, and even though I try to make each note focused on a single idea, they’re better suited for longer, more holistic discussions. Spaced repetition prompts are more suited to laser-focusing on a single detail. A collection of prompts can dissect a complex system, but only once I understand it quite well—and sometimes that understanding can only come through prose writing. The two practices are focused on different outcomes, too: the evergreen notes make ideas available in prose for future writing and thinking; spaced repetition prompts make ideas available in memory for those future tasks. The latter is more ideal—but it’s more costly, and doesn’t work as well when the ideas aren’t so precise.
One useful guiding heuristic has emerged: I write evergreen notes mostly about my own ideas, or to develop the relationship between others’ ideas and my own; I write spaced repetition prompts mostly to internalize others’ ideas, or knowledge about the world.
Of course, the lines here are blurred. When I sit down with a book, I’ll often begin by ensuring I understand what the author means; then I’ll find myself making connections to my own ideas, shifting away from what the author intends for a while. This isn’t a one-way movement: the two modes will often interleave. To support fluidity switching stances, I’ve integrated spaced repetition prompts into my notes, so that I can jot a handful of prompts alongside paragraphs of extended prose.
The integrated note/prompt surface makes my coexistent practices a bit more natural, but I notice there’s still a significant amount of duplicated work. I’ll often use prose writing to figure out exactly what I think an author means by a certain line of argument, then I’ll turn that conclusion into a few spaced repetition prompts. In many cases, this is basically just busywork: writing the prompts is a matter of translating my prose sentences into “prompt-speak”, almost a kind of style transfer. The resulting duplicative text adds line noise to my notes and makes them less usable. Observations like this make me enthusiastic about an “idea-centric” memory system in which items are prose insights-in-context, and practice tasks are adaptively generated from those highlights.
One last collision between note-writing systems and spaced repetition systems is the possibility of somehow placing the notes themselves on a spaced repetition schedule. I was quite enthusiastic about this a few years ago, but I’ve tried many permutations of this idea without compelling results. For instance, perhaps my environment should surface random notes from the past each day, to foster serendipity. Or perhaps, if I have some spark of an idea but I don’t know what to do with it, I should make it reappear a month later. On a couple occasions this has led to generative writing sessions, but the rest of the time it’s just seemed like noise. The key issue here seems to be emotional connection: I get interested in a topic, so I write a note… but when it’s algorithmically resurfaced, I’m probably thinking about something else, and that old note doesn’t seem meaningful. I’ll often find myself returning again and again to old topics, but that rekindled interest doesn’t happen on command.
Future prospects
I’m somewhat baffled by most of the suggestions I’ve seen for new note-writing tools, particularly in the recent age of LLMs. I don’t need my writing assistant to polish the prose in my working notes: I’ll develop those ideas further through the process of writing them into a coherent whole. Basic issues of language, grammar, and syntax are not bottlenecks for me. I don’t need related notes to be automatically resurfaced: search and links have worked fine for me, and I rarely notice myself “losing” notes. I don’t need meeting or book summary generation in my notes: writing a précis is an important part of how I solidify my understanding. I don’t need help shifting the tone of my language around. And so on.
So, where might I benefit from help?
One angle would be to avoid writing these “evergreen notes” at all. If I could develop ideas directly in stream-of-consciousness scratchwork, but they could still meaningfully accumulate over time, maybe I could avoid the overhead involved in creating and maintaining “clean” notes. Latticework gestures in this direction, but I think a tool specialized for this purpose could go further. Ideally, I could ramble about the topic in an unstructured format, but with quick and legible handles back to the “good stuff”. I’d like to be able to incrementally refine those rough ideas over time, without the ceremony of “creating and naming a note”.
Another angle is to notice that I get stuck a lot when writing. Often the approach I’m taking isn’t working, but I don’t have the meta-cognitive wherewithal to step back and choose a different strategy. Worse, I often fail to even notice that I’m in this situation when I’m in it. In the moment, it just feels like aversion and dissatisfaction. What’s interesting here is that if I complain about my stuckness to creative peers, the conversation almost always leads me out of the swamp, usually through utterly familiar suggestions. Is it possible to create a writing assistant which can do something similar, and which could even notice when I’m stuck in this way—without it feeling like Clippy?
While I’m not excited about AI-generated suggestions for connections within my notes, I’m quite interested in suggestions for outside resources I should consult. I notice an enormous advantage that my academic friends have: their environment is full of walking bibliographies. They go to lunch with some other faculty or students, mention what they’re thinking about, and their counterpart says: “Oh, have you seen…?” It’s hard to search for something when you don’t know it exists, or that you should search for it. I’d be grateful for a tool which watches my reading and writing, notices my mind’s grasping frontier, and surfaces keen suggestions.
Comments
Eric 'Siggy'
An additional thought: the relationship you describe between evergreen notes and essays or writing projects (which are all-intensive bursts that tend to replace the former) is similar to the relationship between having a team document work products and experimental results continuously vs. pushing to create a deliverable (ie a presentation or a paper). For example, I like to have teams document all experimental results in a lightweight wiki as we go, which in theory is a staging area for work we will put into deliverables later. In practice, though, creating a deliverable is an all-intensive process of pursing a well-integrated, perfect narrative, and once we have that kind of focus, that crucible experience *replaces* the wiki. I haven’t found the best way to balance the continuous production vs. deliverable push modes, either personally for teams. I’ve gotten pretty good at it for literature review (where it is easy enough to maintain both evergreen notes and a paper draft simultaneously), but not for other parts of the creative/scientific process (like generating plots of data, or refining problems statements and research questions as our understanding of the fog evolves).
Eric 'Siggy'
I appreciate the update on your experience with evergreen notes! I’ve been trying for 3 years or so to implement the idea for myself (in Obsidian), taking inspiration from you. I quickly ran into the same technical debt problem. I’ve settled on a version of the idea that is working very well for me: I have just a handful of “evergreen” pages, of two types: project “home pages” and “knowledge base” or field of study “home pages.” I found quickly that trying to create a dense directed graph of small notes was unwieldy. Instead I use more of a tree: for any idea I want to track, I know where to find the appropriate “home page” with a few clicks. Transient notes (organized by date) always “hang off” some evergreen “home page” they belong to. I find this works very well for literature review (tracking other people’s work—which I do with hierchical lists of hyperlinks to papers grouped by themes and sub-themes (which I am always reorganizing), rather than trying to densely link longer-form paper notes) and project tracking (tracking people’s work internal to my employer of 10,000 people). Pages that are core to a current project (ex. lit review for research areas we will be writing papers on) are always better maintained and less of a “duty” to add things to. Pages that are over ambitious (“track all the LLM work anybody I know is doing”) fall into disuse, and that’s okay. What I love about all this is it gives me a clear action point when I discover a new paper or relevant work by a colleague—a clear Zettl-like filing system for everything I want to track and cluster into project-relevant themes. But my system falls short of the original ambition of evergreen notes: I don’t build much of a hierarchy of my own creative ideas. It’s better for tracking external references in rich ways. But doing that elsewhere is extremely difficult, so I’ve found it useful! Bonus: I’ve gotten 6 or 7 of my project teams in the last couple years to organize all of our shared files (not notes, files!) around the “evergreen” vs “transient” concept (though I rename them into three categories: 1) transient products we create, 2) “knowledge base” products others create, and 3) final products we create). It works well for collaborative file organization, where too many cooks in the kitchen often makes for a mess! As for spaced repetition, I find SRS and evergreen notes to be naturally disjoint. I would never use SRS to organize meeting notes, scratchwork, or a lit review of 200 papers or tracked projects! Likewise, though, notes (evergreen or not) do little to help me really master algorithms or client domain concepts I need to be part of my active vocabulary (that’s where SRS shines). For me the interface between the two is what I call “Learnables:” a kind of transient note where I take notes about a talk, paper, or book in a sort of “first-pass SRS” format. Not just anything I read is learnable (a candidate for a flash card). This list of Learnables becomes a staging area for SRS harvesting, in the form of a markdown checklist. I find this intermediate note format between SRS and traditional notes helps the process of processing information into SRS-worth, high quality nuggets more approachable—especially in firehouse situations like conferences (or visiting a museum), where direct Anki fixation is impossible, but other kinds of notes instantly become stale. Mostly I find this pipeline useful because good cards can’t be made directly from source material in most cases. It’s better to make a list of the “vocab” you want to learn from a text (ex. “Affine hull,” “the Cimmerian invasion of Anatolia,” “the key way Husserl says phenomenology differs from psychology”), and then Google around for additional resources to flesh out or refine the concept into a few good cards. Using transient notes as a staging area gives me natural space to do this postprocessing when time allows. Waiting to do all that card-design activity directly in front of an Anki window creates a bottleneck! It’s been fun: I can collect “learnable” notes far faster than I can create cards. I like watching my collection of Learnables grow around areas that interest me—even if I won’t ever have time to fully Ankify them all.
