Intro

I keep my own task manager, and I’d been itching to rewrite it as file-based (one task per file + an index, editable directly in vim, greppable, git-able). While digging into that, a thought showed up:

I keep all my conversations and thinking in a Markdown vault anyway — isn’t it wasteful that task management is a separate system? If tasks were just vault notes too, everything would be in one greppable place, which is convenient even for the AI.

The logic holds. And yet it felt deeply wrong before I’d even tried it. This post is what happened when I tried to put that wrongness into words — and ended up dragging the question of where blog drafts live into it, and rebuilding the vault.

Knowledge and tasks split along the direction of fit

The discomfort finally had a name: direction of fit (Anscombe / Searle). Which side — mind or world — is authoritative, and which one do you bring into line with the other?

• Knowledge = belief. Mind → world. The world is authoritative; you fit your description to it. When they diverge, you fix the description (the knowledge). Evaluated true / false. There is no “done”; it accretes as you revise. Durable.
• Task = intention. World → mind. The intention is authoritative; you bring the world into line with it. When they diverge (incomplete), you fix the world (not the intention). Evaluated done / undone. It disappears once achieved. Stateful.

That yields a one-shot test:

Before saving anything, ask: “Is this evaluated true/false, or done/undone?” true/false → knowledge (vault). done/undone → task (a dedicated store).

Mixing a high-churn state machine (tasks) into a knowledge graph is like binding a TODO list into an encyclopedia. The git history gets muddied by done/undone toggles, and links to finished tasks break. What I’d felt as “separation is wasteful” was really just the overhead of running two systems — and solving that by merging breaks things. The efficiency I actually wanted comes from a link, not a dirtying merge.

The ick-spectrum — where blog comes in

The interesting part is that the ickiness isn’t binary; it’s a continuum. Things line up by which evaluation axes (true/false / done/undone) ride along:

• Permanent note … true/false only → clean as-is (pure knowledge)
• Blog draft … true/false + done/undone → partially icky (hybrid)
• Pure task … done/undone only → maximally icky (lives outside the vault, full stop)

A blog draft, which looks unrelated to tasks, sits right in the middle of the same ruler. As content it’s true/false (is the post any good?), but it also carries a publish lifecycle (draft/published/rejected) — a done/undone. So it’s not as clean as pure knowledge.

And my vault had exactly that middle thing sitting in it, unaddressed: an 05_Blog/ folder (draft/ published/ rejected/). Only once I’d named the task case did I see it: “this is a mild case of the same disease.”

What I actually did — clearing the middle thing into a pipe

Tasks already live in my dedicated tool. What was left over was the spectrum’s middle case — the blog drafts — so that’s what I cleaned up.

• Deleted 05_Blog/ entirely. Posts now live only in the blog repo.
• Stripped the blog lifecycle conventions out of the vault README, leaving just a gather convention: “bundle the relevant permanent notes.”
• Built a stateless pipe from knowledge to published artifact: gather material from the vault (source), shape the post in a dialogue, mask it, and emit into the blog repo (artifact). No draft artifact is kept anywhere — the draft is the conversation itself.

This is where a durability axis earns its keep (source / cache / artifact). The test: “if I delete this, do I lose something irreversibly?” Vault knowledge is source; a published post is an artifact (a derivative that, once emitted, has its own existence outside); a draft is neither — it’s in-progress working state. So it isn’t persisted.

Outcome

• The vault is back to a clean “true/false only” knowledge base; the task-smell of lifecycle state is gone.
• Deciding where something goes became mechanical: just ask “true/false or done/undone?” and “irreversible if deleted?” before saving.
• Incidentally, this very post is the first thing that came out through that new pipe.

Takeaways

• Knowledge and tasks split along the direction of fit: true/false (knowledge, vault) vs done/undone (tasks, a dedicated store).
• The ickiness is a continuum: pure knowledge → blog draft (hybrid) → pure task. Blog sits in the middle of it.
• A draft is neither knowledge nor task — it’s in-progress state built from a source. Don’t persist it.
• Separation isn’t wasteful. You get proximity from a link, not a merge.

See also

• A plugin holds WHEN, the vault holds HOW — the same “separate responsibilities to stay drift-proof” principle, applied elsewhere
• The real purpose of a Zettelkasten was output — the tug-of-war of over-tending the vault until the means becomes the end