Skip to content

Memex Operating System

Memex runs on a simple loop:

Capture → Classify → Link → Pressure-test → Act → Learn

This is the complete operating loop for Memex Mark 1 Operating Core. Every iteration strengthens the system, tightening the relationship between what you capture and what you decide.

The harness layer adds one more operating loop around the knowledge loop:

Prepare -> Execute -> Trace -> Verify -> Govern -> Learn

1. Capture

Capture promising raw material quickly, without over-processing.

Examples:

  • book passages
  • article excerpts
  • meeting insights
  • founder/investor/operator lessons
  • market signals
  • regulatory observations
  • strategic questions

Capture belongs in content/sources/.

2. Classify

Turn sources into memex items with clear type and structure.

A classified item should answer:

  • What is the core thesis?
  • Why does it matter?
  • What are the key ideas?
  • What mental model does it strengthen or challenge?
  • Where might this apply?
  • What question should we keep asking?

Classification belongs in content/items/ or the appropriate folder based on type.

Connect new ideas to existing themes.

Useful connection prompts:

  • What does this remind us of?
  • Which assumption does this challenge?
  • What second-order consequence follows?
  • Which decision could this improve?
  • Which strategic theme does this reinforce?
  • What would inversion say?

Future maps will live in content/maps/.

4. Pressure-test

The system is only valuable if claims survive scrutiny.

Pressure-test notes by:

  • Challenging assumptions with counter-evidence
  • Running claims through the Skeptic agent
  • Checking predictions against outcomes
  • Comparing beliefs to reality
  • Seeking disconfirming evidence

A claim that has never been tested is not yet knowledge.

5. Act

Apply notes to real decisions:

  • strategy memos
  • investment or product decisions
  • regulatory positioning
  • market expansion questions
  • leadership and hiring principles
  • personal learning plans

A note that never changes thinking is probably just storage.

6. Learn

Feed outcomes back into the system. When a decision resolves:

  • Update the prediction ledger with actual outcomes
  • Revise probabilities based on new evidence
  • Archive beliefs that no longer hold
  • Surface patterns that repeat

This closes the loop. What you learn becomes what you capture next.

Harness operating modes

Every agent run trades off cost, quality, and speed. Use the lightest mode that is appropriate for the risk.

Mode Use when Required behavior
Fast Read-only lookup or low-risk synthesis Search, cite, answer, no durable writes unless requested
Standard Normal ingest, docs, or repo edits Retrieve, cite, update files, run relevant validation
Strict Private data, public release, governance-sensitive work, or long-running automation Run readiness checks, trace important actions, validate, preserve state, require human approval where policy says so

Default to Standard for framework changes. Use Strict for public release, private-vault exposure, permission expansion, or external actions.

Periodic Review

Review periodically to compound the system.

Suggested cadence:

  • Weekly: clean new captures and classify items.
  • Monthly: merge repeated tags, update maps, identify emerging themes, calibrate predictions.
  • Quarterly: ask which ideas changed decisions or deserved deletion.

Quality bar

A memex item is ready when it contains at least one of:

  • a clear thesis
  • a reusable mental model
  • a decision-relevant insight
  • a high-quality question
  • a cross-domain connection

If it has none of these, keep it as a source or discard it.