Knowledge Base as Thinking Partner
Knowledge Base as Thinking Partner
The knowledge base should help decide what to think about next and still hold what has already been learned. Did this page make the next action clearer? A no on that question is the system drifting toward storage.
Archive or partner
An archive stores finished material. A thinking partner changes the next move. The page is not the endpoint. The page is a thinking surface.
If the test fails, the system is drifting toward storage. The partner’s kinds of help are pattern-finding, gap-finding, frame-testing, question-generation, and turning a vague interest into concrete work.
Every interaction leaves at least one of five outputs. A sharper question. A better frame. A visible gap. A next action. A reusable synthesis. A session that produces none of these retrieved well and still failed the test.
The mechanism is two beats. The current mental model is written down. Related pages then pressure-test it. A partner that only retrieves is an archive with better search.
The loop is:
interest → question → relevant pages → pattern or contradiction → synthesis → next action → updated pageThe last arrow is load-bearing. Without the write-back the loop is browsing.
The knowledge base compounds when old pages start changing how new problems are approached, not only by adding sources.
| Archive asks | Partner asks |
|---|---|
| Where should this go | What does this change |
| What did this source say | What can I do differently now |
| What pages exist | What pages should talk to each other |
| What do I know | What am I missing |
Archive mode is still useful. It keeps the system legible. Thinking-partner mode is where the system becomes alive.
Five moves
Start from an active pressure, not a random topic. The system serves current judgment, not only future retrieval. The prompts that find that pressure: what is being decided; where the work is stuck; what keeps recurring; what feels important but under-formed; what would make this easier to act on.
Pull nearby pages. Do not start from scratch. Graph, index, backlinks, and the obvious links are enough. The starter packs are doors into this wiki, not summaries of those pages. Categories stay stable. Specific doors follow the live names.
Learning: Prestudy, BHS, and SIR: Turning Information into Usable Structure is the encoding-and-retrieval synthesis. First Principles of Learning is the mechanism layer. Are You Learning, or Just Using Techniques is the diagnostic that asks whether the method is doing any work. Knowledge Mastery: From Recognition to Usable Knowledge is the ladder from recognition to use.
Execution: Attention Management: Preserving Flow is the flow-preservation door. Focus Management: How to Enter & Recover Inside a Work Block is entry and recovery inside a block. Procrastination: a System Problem is delay as a system, not a character flaw. Decisional Delays is the stall that looks like thinking.
Judgment: Applied Critical Thinking: Testing Frames is how a frame gets tested. Decision Making is the decision page. How to Communicate Truth Into Someone Else’s Frame is the transfer problem. Suicidal Empathy is empathy that erases a necessary line.
Building: Agentic Engineering is how agents get built here. Understanding Bottleneck is the comprehension stall. Interleaving for Complex Problem Solving is mixed practice on hard work.
The useful part is usually the tension between pages, not the page itself. Look for a concept with no operating model; two pages that imply different actions; a repeated failure mode; a missing theory/practice bridge; a technique that does not change behavior; an interest with no next action. Pages that came from that tension: Balancing Multiple Interests: Breadth v Focus is the breadth-versus-focus collision. The Shortcut Problem is the cheap path that skips the encoding. The Technique Is Only as Good as the Thinking It Produces is method that looks right and changes nothing. How to Unlearn Old or Bad Habits Efficiently is the unlearning problem.
Convert the insight into a handle. A handle is a short name, question, or rule that can be picked up later when the situation returns. Title, question, checklist, contrast, next exercise, decision rule, page split, journal prompt. The handle is what makes the idea reusable. A good handle lets the system call the idea back at the moment it matters.
Worked conversions:
- “better study habits” → Prestudy, Bear Hunter System, Spaced Interleaved Retrieval
- “too many interests” → Balancing Multiple Interests: Breadth v Focus, Priority 0+1
- “consuming without changing” → Are You Thinking, or Just Consuming?
- “method looks right but nothing improves” → The Technique Is Only as Good as the Thinking It Produces
File the result into the smallest useful destination. Write-back is a move, not a courtesy.
| Kind of result | Destination |
|---|---|
| Passing thought | Journal |
| Active question | Journal, What’s On My Mind |
| Current working area | Journal, Current Focus |
| Durable operating model | wiki |
| Personal plan | private Goals or Skills |
| Public argument | blog draft |
| System change | README, AGENTS, or a workflow page |
A page is not created just because an idea exists. A page is created when it will be reused, linked, practiced, or argued with — the same candidate test Wiki Breakdown Pass applies when a new page is on the table.
When it is working
Used well, it feels like returning to a room where the important threads are already laid out. Old pages are changing how new problems get approached. A session that produced no write-back failed the test, however much was read. This is not another feed; the mechanism that turns a vault into one lives on Social Media - Curvilinear Design & the Theft of Time.
One session, the failures
When the system feels passive, one session is enough. The goal is not to process the whole vault. The goal is to make one part more usable.
- One live question.
- Read three to six related pages. The bound is a house limit, not a finding.
- Write the current answer in plain language.
- Name the tension.
- Pick a destination.
- Add one link, one next action, or one open question.
| Failure | Repair |
|---|---|
| Knowledge hoarding | Ask what the page changes |
| Index addiction | Start from a live problem |
| Link decoration | Link only when the relationship matters |
| Permanent synthesis | Use Journal for temporary thinking |
| Private/public confusion | Keep private Goals, Skills, and Kolbs local unless they are intentionally publishable |
| Agent overproduction | Ask for operating model, causal chain, failure modes, practical feel, and implications |
| Archive drift | End each session with a question or an action |
Kolbs, in this vault: after an attempt, what happened, how it felt and why, what rule that suggests, what changes next time.
Signs, weekly, the price
Good signs: the next question is obvious; old pages explain new problems; titles have become handles; interests are alive but bounded; the Journal reflects current attention; the index is stable enough; outputs are shorter because context is already held.
Warning signs: every session starts from zero; pages are polished but unused; the graph is interesting but not actionable; new sources do not update old pages; Journal questions sit unchanged for weeks; the system is impressive and inert.
Weekly questions, asked once: what page changed how the week was acted; what question kept returning; what thread is stale; what should be synthesized next; what belongs in a blog post. The weekly review is not a productivity ritual for its own sake.
A partner that never writes back is an archive. Starter packs rot as pages are renamed; the categories stay and the doors follow the live names. Artifact production without a live question is the agent-overproduction failure. A session costs three to six pages plus a write-back, not a tour. Two sessions that start from zero, a week of Journal questions unchanged, or a weekly review that produces no next action are quit signals. One session leaves a destination, and the test can be answered yes or no.
The same test, after something has been filed: did this page make the next action clearer?
Links into the knowledge base
- LLM Knowledge Systems — the three-layer pattern this workflow sits on.
- Raw to Wiki Compilation — how sources become pages.
- Question Answering Against a Wiki — how a question is answered from the wiki first.
- Wiki Health Checks — the lint sibling.
- Metacognition: The Control Layer — the control layer the weekly review is a special case of.
- Agentic Engineering — building-starter door; the questions this page starts from.
- Journal — temporary thinking; the destination for a passing thought, an active question, and current focus.
Open Questions
Starter packs will go stale during regen — how should a later pass refresh the doors without rewriting the moves?
When is archive mode the right mode for a stretch of work, rather than a drift to correct?
Sources
- LLM Knowledge Systems — vault pattern this workflow sits on.
- AGENTS — house operating rules for ingest, query, and write-back.
- Sönke Ahrens, How to Take Smart Notes (2017); Vannevar Bush, “As We May Think” (1945); Douglas Engelbart (1962) — the popular “notes as thinking” lineage. Named here only. Not required reading. Nothing in them replaces the five moves or the archive/partner table.