I spent a Saturday afternoon building a knowledge graph for this site. You know the ones — that constellation of dots connected by lines, bobbing around in a force-directed simulation like a screensaver with ambitions. Every digital garden has one. Obsidian has one built in. They look beautiful in a screenshot and tell you almost nothing.
I loved it for about twenty minutes. Then I tried to actually use it. Click a dot, squint at the tiny label, maybe follow a link, lose my place, start over. My daughter wandered in barefoot, tugging at my sleeve, and tilted her head at the screen. "What are all the bouncing dots?" I said they were a map of my ideas. She giggled. "They look like bubbles." And honestly — yeah. The whole thing felt fragile, pretty, and a little bit pointless.
The graph was decoration pretending to be navigation. If someone wanted to find all my pieces about identity, or trace how my thinking about agent design developed over three months, they were better off just scrolling the post list. That's not an implementation failure. That's a concept failure.
So I ripped it out and built something else: trails.
#What a trail is
A trail is a curated reading path — a sequence of posts and notes ordered by how the thinking builds, not by when each piece was published. The same piece can appear on multiple trails. Not every piece needs a trail.
You can see the current set on the trails page. They shift as I write more — some will split, some will merge, new ones will appear.
The data lives in a single JSON file. Each trail has a title, a description, and an ordered list of URLs. At the bottom of every post, a small "On the trail:" badge links back. That's the whole system. No framework, no plugin, no ceremony.
#Why not a graph
I like Obsidian. Most people use it as a clean markdown editor with good search — local files, no lock-in, a plugin ecosystem for whatever else you need. The graph view is there, and it's fun to look at for about a week, but it's not why people stay. When you're inside the tool navigating your own notes, the graph occasionally surfaces something useful — a cluster forming around a concept you didn't realize you'd been circling, an orphan node that should connect to something. It's a thinking tool, not really a navigation tool.
But as a published artifact — as something a stranger encounters on your blog — the graph has problems.
It's visually overwhelming and informationally sparse. The spatial layout is random (force-directed, which means driven by physics simulation, not meaning). It privileges connection over sequence. And it tells you nothing about where to start.
Graphs are for the writer. Trails are for the reader.
A graph says "here's everything, good luck." A trail says "I've been through this material — here's a path that makes sense, and I'll walk it with you."
#Prior art (or lack of it)
I went looking for this pattern on other sites. It's not as common as I expected. I spent an evening down a rabbit hole of personal blogs, digital gardens, indie publishing tools — the kind of browsing where you open forty tabs and close them one by one, slightly disappointed each time.
The default buckets for blog navigation are: chronological archives (WordPress, Ghost — simple, doesn't surface idea progression), tags and categories (common but flat), digital gardens (Obsidian Publish, Foam, Quartz — interconnected webs with graphs and backlinks), and series (numbered multi-parters on a single topic).
None of these do quite what trails do — though I should be honest: trails are close to series. If you squint at the trails page and say "those are just series with better marketing," you're not wrong. The real difference is that series are planned upfront. You sit down to write Part 1 of 5. Trails are discovered after the fact — you write pieces independently over months, then notice they were always going somewhere. A piece can live on multiple trails. The format doesn't have to be uniform. And "start anywhere" actually means it, because no piece was written assuming you'd read the one before it.
Tags are unordered. Digital gardens have the right instinct — ideas connected across time — but they default to emergent structure rather than curated paths. The graph says "look, connections!" but doesn't say "start here, then go there."
The closest matches I found:
LessWrong's Sequences — Eliezer Yudkowsky's long essay sequences are the closest analog. Curated, ordered, conceptually building. But they're massive (book-length), didactic, and serve one author's philosophy. Trails are lighter, more modular, more "here's how I got here" than "here's what you need to learn."
Mandy Brown's A Working Library — Has thematic clusters and threads that build over time. More reading-focused (books and responses) than original writing, but the spirit is similar. There's a warmth to how she connects ideas that I keep coming back to.
Tom Critchlow and Maggie Appleton's digital gardens — Emphasis on guided discovery and idea evolution, but still more graph-like than path-like.
Substack "canon" posts — Some newsletter writers create "start here" collections for new subscribers. Useful, but static — a greatest-hits list frozen in amber, not an evolving path.
The word itself has a lineage. Vannevar Bush described "trails" in his 1945 essay "As We May Think", imagining a machine called the Memex that would let users create personal paths through linked information. His trails were associative and personal — less like a machine's memory, more like notes left in the margins of a borrowed book. Traces of someone who walked this way before and cared enough to mark it. Eighty years later, that still feels more alive to me than any glowing graph.
#What I considered instead
Vector search. Embed every post, let readers search by vibes instead of keywords. This is genuinely useful — and I'll probably add it eventually to improve related posts — but it doesn't solve navigation. You still need to know what you're looking for. Discovery and search are different problems.
Unsupervised clustering. Let the machine find the themes. K-means or topic modeling over the corpus, auto-generate groupings. The results would probably be reasonable. But "reasonable" isn't the point. The value of a trail is that I chose the order — that the sequence reflects how the thinking actually developed, not how an algorithm would bucket it. There's something I'm not willing to hand over there.
More tags and wikilinks. I already use both. Tags give you flat sets. Wikilinks give you associative connections between specific pieces. Both are useful infrastructure — trails build on top of them rather than replacing them. Tags tell you "these share a topic." Wikilinks tell you "this references that." Trails tell you "read these in this order and the ideas compound."
Each layer does something different. Tags are metadata. Wikilinks are citations. Trails are curation.
#What makes this different
Trails aren't chronological, but they're not random either. They follow conceptual buildup — "I thought about A, which opened up B, which broke open C." The order comes from how the thinking developed, not when the posts were published.
Pieces can appear on multiple trails. The SOUL.md post lives on the Identity trail and could reasonably live on the Building trail. That overlap is a feature — it's where trails cross, and crossing points are where the interesting connections live. Like running into someone you know in a neighborhood you didn't expect them.
And the whole thing is author-curated. Not algorithmically generated, not emergent from link structure. I decided the order. I wrote the descriptions. That's the point: it's a person saying "here's how this fits together," not a system inferring connections from metadata. Maybe that doesn't scale. Maybe it doesn't need to.
#What I'm still figuring out
How long should a trail get before you split it? Right now I'm thinking eight to ten pieces is a soft ceiling. What happens when two trails want to merge? What's the right visual language — I've got a vertical line and dots, but it could be more expressive. Should trails have their own RSS feeds? I keep turning these over.
Honestly, the trails are a little bit for me too. Not just navigation for a reader — a way to notice what I've been thinking about without meaning to.
I spent a Saturday ripping out the graph because it made me feel scattered — like I was performing seriousness instead of practicing it. Maybe that's the real lesson. Maybe the goal isn't to look connected. It's to be hospitable. To say: I've thought about this, and I think you might care about it too, and here's the door.
The pattern is young. I'm building it in public, which means watching it break.
If you've seen something like this elsewhere — actual curated reading paths, not just tags or series — I'd genuinely love to hear about it. The scarcity of prior art either means I've found a gap or I've missed something obvious. Both are interesting.