Inside ichi · 04

Knowledge Architecture

A pile of notes is not knowledge. Knowledge is what you can ask, connect, and trust to be current.

Every part of ichi runs on one body of knowledge: a single vault of plain-text notes. But a vault of files is only raw material. The architecture is what turns it into something you can think with.

The vault, and why it is plain text

Everything is Markdown in a local, version-controlled vault. No database owns the knowledge; the files are the knowledge. This is a deliberate constraint (see the Architecture Decisions): plain text is portable, diffable, and readable in thirty years. A note written today should outlive every tool used to write it.

From files to intelligence

Four capabilities sit on top of the files:

  • Retrieval. Ask a question in plain language; the system finds the notes that bear on it and synthesises a cited answer. This is the difference between searching a vault and asking it.
  • Connection. The same idea appears in a note on markets, a note on biology, and a note on code. Surfacing that link — that these are the same shape — is where most of the value hides.
  • Patterns. Recurring structures are named and tracked, so a pattern found once can be recognised the next time it appears in an unrelated domain.
  • Freshness. Knowledge has a half-life. The system flags notes that have gone stale, because a confident answer built on an out-of-date note is worse than no answer.
flowchart LR V["Vault — plain-text notes"] --> R["Retrieval — ask, don't search"] V --> C["Connection — cross-domain links"] V --> P["Patterns — named & tracked"] V --> F["Freshness — flag the stale"] R --> A["A cited answer"]

The editorial pipeline

Knowledge that is worth sharing enters a pipeline: intake, analysis, drafting, review, and — only past a human gate — publication. The published essays of WkndPrjct are the visible end of this pipeline. Most of the vault never becomes public; it is the private substrate the public work grows from.

The learning loop

The most important property is that the system learns from its own operation. A failure becomes a documented lesson. A lesson changes a template, a checklist, or a default. The next time the same situation arises, the correction is already built in. Knowledge that does not change how the system behaves is just storage. Knowledge that closes the loop is memory.

The test of this architecture is simple: can the organisation answer a question that no single part of it was told the answer to? When the connections are right, it can — because the knowledge was never in one place. It was in the structure between the notes.