Why the vault exists

// the problem

A result survives more easily than its reason.

Work leaves a scattered record: conversations, changes, experiments and decisions. I may remember the result while losing the reason for it, the dead end that mattered or the check that settled it.

When a similar problem comes back, I want to start from that experience. An agent helping me should be able to find it too, without loading the whole history or guessing what I believe.

// the purpose

Turn experience into useful context.

The vault connects what I built, how I approached it, why I chose a path and what the work taught me. A short account provides the way in. The sources remain available when a question needs more depth.

A small visible surface can have a lot beneath it. Start with an overview, follow a group, open a topic, then read the work behind it. Each step should earn the next layer of detail.

That is why the public landing page stays short. The guides hold the system explanation, and the actual private records remain outside this website.

// design principles

A record I can explain and correct.

• Keep ownership clear. Authored context and generated accounts have different owners. The model can propose an account; an interpretation must retain its attribution.
• Keep a route to evidence. A useful summary leads back to the context behind it. Reaching a source still requires reading it before relying on its claim.
• Show uncertainty. Unknown facts, ambiguous names and unfinished references stay visible as work to review.
• Make repetition ordinary. Explicit schemas, bounded operations and consistent checks make the next run inspectable.
• Use depth deliberately. Give a reader or agent the context the task needs, with links for deeper discovery.

// current direction

Measure whether a later question gets a better start.

Source collection and page generation, authored/generated boundaries, versioned metadata, scoped rules and privacy checks form the foundation. Connected navigation, identity resolution, lexical search, reference maintenance and component diagrams have been implemented; the newer wiki work remains under review.

Cached summaries of discovered themes and cross-project agreement or conflict suggestions are further work. They need bounded inputs, source trails and review before becoming trusted guidance. Code being present is separate from a host running it.

The measure I care about is whether a later question reaches useful, inspectable context. More notes and more links help when they improve that outcome.

Read the origin story: “A second brain built from my chats” →