The case against tagging your notes

You probably haven’t tagged enough, and that’s fine

If you’re reading this, you likely have a vault full of notes where the tagging is inconsistent at best. Maybe you started with a system, kept it up for a few months, and quietly stopped. Maybe you never started. You’ve got years of writing sitting in folders organized by date, and a low-grade guilt that you should have been more disciplined about metadata.

I want to make the case that your vault is more retrievable than you think — and that the energy you’d spend fixing your tags is better spent writing more.

What I got wrong building retrieval the first time

The first version of Enzyme did what most RAG systems do: chunk documents, embed them, match against a query. It worked. The results were accurate. And something about them felt dead.

Users would get back the right passage from the right note, and feel paralyzed. The system answered the question but didn’t move anyone toward something they could do with the answer. Accurate retrieval that leaves you feeling like you didn’t do enough — that’s a system optimized for the wrong thing.

The problem wasn’t precision. It was that chunked retrieval treats your vault like a database. You ask a question, it returns a row. But that’s not how anyone relates to their own accumulated thinking. You don’t want the answer — you want to be reminded of how you were thinking when you wrote it. You want the surrounding context, the adjacent ideas, the thread you were pulling on. The note is a point on a trail, and the trail is what matters.

That experience is what pushed me toward a different architecture: one where tags, links, and folders become sites for generating thematic handles — questions and tensions that probe what your vault is thinking about — rather than filters for narrowing results. Search runs through those thematic handles, not through your metadata. Your tag #decision-making contributes to the context that generates a catalyst like “What does it look like to commit before you have full information?” But the search itself never looks at the tag.

What happens in a vault with no tags at all

I’ve run Enzyme on a vault with zero tags, zero wikilinks, and no structure beyond daily dates. Three and a half years of plain-text journaling. Daily entries with a simple template: checklist, brain dump, course of actions, wins. The person who built it dropped tags early because they interrupted his writing speed. He couldn’t journal at pace if he had to stop and categorize.

Without tags or links, the system falls back to folder entities — whatever structure exists in the file paths. Catalysts still generate. Search still works. The architecture adapts because it was never dependent on tags being present.

The result that surprised me: the owner had been writing about entrepreneurship — the energy of building something, early-stage friction. Enzyme surfaced a journal entry from a hiking trip two and a half years back. The words he used to describe navigating unfamiliar terrain rhymed with how he’d been writing about navigating early business decisions. No tag connected those entries. The connection existed in the language, and temporal weighting of recent entries created the lens that found it.

He remembered the trip. He didn’t remember writing about it in terms that echoed what he was thinking now.

Tags encode your past model of what matters

When you tag a note #creativity, you’re betting that “creativity” is the useful axis for that note later. Maybe it is. But six months from now, when you’re thinking about risk tolerance, that note about creative process — which is really about tolerating ambiguity — stays filed under a category that doesn’t match your current question.

This compounds. A mature tag hierarchy becomes a theory about how your knowledge is organized. The theory hardens. Autocomplete suggests existing tags. New tags feel like overhead. The categories that don’t exist yet — the ones that would capture what you’re actually thinking about now — never get created.

Enzyme regenerates its thematic handles based on temporal weighting. What you’ve been writing this week becomes the lens for the whole vault. A note tagged #creativity three years ago can surface under a completely different thematic handle, because the handle was generated from your current thinking, not your past taxonomy.

Tags are useful — just not load-bearing

I still use tags. They’re good as lightweight clustering signals. A vault with tags gives the system more entity sources, which means more granular thematic handles, which means richer retrieval. Tags help.

But they don’t need to be perfect. They don’t need to be consistent. They don’t need a hierarchy. The system reads context around the tag, not the tag itself. A tag you used loosely and dropped after three months still generates useful catalysts from the passages where it appeared.

The guilt about inconsistent tagging, the friction of categorizing in the moment, the hours spent designing tag systems — that effort is aimed at a retrieval mechanism that isn’t doing what you think it’s doing. Your tags are hints. They’re not the search index.

Write more, categorize less

The vaults where I’ve seen the most interesting retrieval aren’t the most organized. They’re the ones with the most raw writing. Daily journals, voice memo transcriptions, brain dumps where someone was thinking out loud.

Structure helps for manual retrieval — following your own links, filtering by tags you remember. But when retrieval uses temporal reasoning and thematic search, the value of your vault scales with the volume and honesty of the writing, not the precision of the metadata.

The case against tagging isn’t that tags are bad. It’s that they’ve been carrying weight they were never meant to bear. Let them be lightweight. Let them be messy. Write more instead. The retrieval system can do the work that your tags were pretending to do.

Try it

Enzyme indexes Obsidian vaults using temporal decay and catalyst-mediated semantic search. Tags enhance it but aren’t required. If you’ve been writing without structure and wondering whether any of it is retrievable — it is.