Skip to content

Governance

Operating Standard

Memex is a framework for institutional memory. Treat writes to real vaults as durable acts.

Rules:

  • Search before answering.
  • Cite before claiming.
  • Add provenance before trusting.
  • Preserve raw sources.
  • Log durable writes.
  • Never publish private vault content by accident.

Data Classification

The Memex Mark 1 Operating Core uses a six-grade classification system. Every note is assigned a grade. Every output respects it. Human review is required at every boundary crossing.

The Six Grades

Grade Safe for Example content
Personal private Vault only Health records, family notes, personal journal
Company private internal Team members Strategy memos, financial models, org charts, internal postmortems
Investor-ready Fundraising context Pitch decks, term sheets, cap tables, growth metrics
Regulator-safe Compliance audit Audited financial statements, regulatory filings, compliance documentation
Public content Anyone Blog posts, published papers, open-source docs, conference talks
Group chat safe Low-stakes communication Scheduling, logistics, public links, meeting reminders

Why Six Grades?

The original four-class model (Public / Internal / Confidential / Sensitive) collapsed too many distinctions. Investor materials and regulatory filings have different disclosure rules. Published content and group-chat messages have different review requirements. The six-grade model separates these cases explicitly.

See Firewall & Data Classification for the complete policy.

Boundary Crossing Rules

A boundary crossing happens when a note classified as one grade is included in an output targeted at a lower-trust audience. Every boundary crossing requires human approval.

Example: An essay draft (Public content) citing an internal strategy memo (Company private internal) triggers a boundary-crossing review. The operator must either approve specific quotes with redaction or exclude the private note entirely.

Provenance Rules

Every important claim should be traceable:

provenance:
  - date: '2026-05-28'
    claim: 'The claim being made.'
    source: '[[sources/source-slug]]'
    confidence: high

Confidence levels:

  • high: primary source or direct record
  • medium: credible secondary source
  • low: inference or weak source

Write Rules

Agents may:

  • create draft pages
  • update working notes
  • add provenance
  • add backlinks
  • generate briefs
  • propose merges

Agents must not:

  • rewrite raw sources
  • delete pages without redirects
  • publish externally without approval
  • invent facts
  • silently overwrite human edits
  • commit secrets

Review Rules

Use Git for review:

git status
git diff
git add .
git commit -m "Update brain notes"

For team environments:

  • use pull requests
  • protect main branch
  • require review for schema changes
  • allow agent-generated branches

External Sharing

Before sharing anything outside the team:

  1. confirm no private entity pages are included
  2. scan for secrets
  3. remove raw transcripts
  4. replace real examples with synthetic ones
  5. publish only the framework repo, never an unsanitized private vault

Basic scan:

rg -i "api[_-]?key|secret|password|token|private|confidential|sk-" .

No automated scan replaces judgment.