Vault
analysis

The Positioning Vault Pattern

Created

The Positioning Vault Pattern

Related: _index, grand-scheme-advice-and-unexplored-ideas Informs: ACTIVE

The trigger

I started this folder on 2026-03-22 with no particular plan beyond "keep the research." Three weeks in, I asked Claude to build a running list of active projects, which turned into a broader question: is this scaffolding a thing? Is anyone else doing this?

Turns out yes. And the pattern has no agreed name.

The architecture

~/Projects/research/ is one folder wearing three hats at once — knowledge vault, strategy workbench, and meta-orchestrator over sibling codebases.

The tree

~/Projects/research/
├── CLAUDE.md                  # Conventions — tags, typed links, tone, slash commands
├── ACTIVE.md                  # Running list of what's in-flight
├── docs/
│   ├── _index.md              # Curated thread map of how docs connect
│   └── *.md                   # Research docs: frontmatter + typed links + body
├── conversations/             # Claude.ai exports processed into markdown
├── places/                    # Seattleite guide (deterministic scripts + MD entries)
├── tools/
│   ├── forecast/              # forecast_lab — macro regime detection
│   └── portfolio/             # portfolio_lab — rebalance tracking
├── worktrees/                 # Blog draft branches (managed by tech-blog repo)
├── scripts/                   # Processing utilities
├── .claude/commands/          # Slash command definitions (8)
└── .obsidian/                 # Vault config (committed)

And the paired memory layer, stored outside the folder in Claude's project-scoped memory directory:

~/.claude/projects/-Users-ryuhei-Projects-research/memory/
├── MEMORY.md                  # One-line-per-entry index
├── user_*.md                  # 13 files — who I am, how I work
├── feedback_*.md              # 8 files — corrections and validated approaches
└── project_*.md               # 6 files — active projects

The conventions

Every doc in docs/ has:

Every memory file has a type (user / feedback / project / reference), a description that doubles as a relevance signal for future sessions, and a one-line entry in MEMORY.md.

The slash commands

Eight vault-specific verbs live in .claude/commands/:

These make the vault act more like a REPL than a filing cabinet.

The landscape

The meta-question: who else is doing this, and do they call it anything?

Short answer — it's emerging in several places with different weights and no consolidated name.

Terms in play: LLM Wiki, agentic knowledge management (AKM), knowledge OS, agent-native PKM, spine pattern. None has won.

What I'm actually running

Three things collapsed into one folder that most public templates keep separate:

  1. Knowledge compaction — memory directory + MEMORY.md index. The Karpathy move.
  2. Essay and strategy development — docs/ with a dense typed graph. The Appleton move.
  3. Code and artifact production — tools/ with committed analysis code, worktrees/ for blog drafts, sibling projects linked via Led-to:. The Soporan move.

Nobody public is clearly doing all three in one folder with conventions this formal. The typed-link discipline (Related / Builds-on / Informs / Led-to) is the most absent part of the public conversation — closer to Andy Matuschak's evergreen-notes habits than to the LLM-Wiki crowd.

The usage pattern

From a 60-day audit of the vault and the memory directory:

Dimension Shape
Docs 27 in 22 days. Seed burst of 7 on Mar 22, then steady ~1/day
Type mix 14 research, 7 analysis, 3 reference, 2 plan, 1 index
Graph density 22 docs with Related, 21 Informs, 17 Builds-on, 3 Led-to
Tag head personal (19), strategy (9), self-reflection (9), market-analysis (6)
Memory 28 files: 13 user, 8 feedback, 6 project, 0 reference
Commits (60d) 50 total — 15 portfolio_lab, 8+ forecast_lab dominant
Slash commands 8, all local to this repo

Two takeaways:

Where it leaks

  1. Led-to is underused. Only 3 docs have it, and one was added during this session. portfolio_lab and forecast_lab both exist because of earlier docs, but nothing in the graph shows that. Every worktrees/draft-* should have a Led-to: from its source research doc; every tools/ subfolder should too.
  2. reference memory type is defined but never written. CLAUDE.md has four types; only three are used. Either drop it or start writing external-system pointers (Linear projects, Grafana boards, Slack channels).
  3. No "completed" signal on docs. Every research doc is equal in the graph regardless of whether the thesis still holds, the decision has been made, or the topic has moved on. A status: active | settled | stale frontmatter field would make this visible without losing the archive.

The signature moves

What makes this setup unusual vs. the public landscape:

What I'd call it

If I had to put a name on it: positioning vault. Not a wiki, not a second brain, not a PKM — those are archival framings. This is infrastructure for a specific person's active decisions.

Research feeds decisions. Some decisions spawn code or drafts. Memory holds ground truth about the person using it. Slash commands make the vault operable as a tool rather than passive storage. The whole thing has a direction — the person it serves is trying to get somewhere specific, and the folder is shaped around that trajectory.

The public conversation is still mostly about knowledge compaction or code coordination in isolation. The interesting thing is that you can collapse all of it into one folder if you're willing to enforce conventions.

Where this goes next

The leak fixes are small and mechanical:

The broader interesting piece: the "folder over multiple codebases" pattern is already in play at work — I use a version of it professionally — and the typed-link discipline is portable. Non-PKM teams would use it to track decisions → artifacts rather than ideas → essays. That's probably the blog post: the architecture here is concrete enough to demonstrate and general enough to adapt.

Backlinks