guide to agentprivacy
Browse collections
โœจVisualise
Connect with Star
Your VTA, your chosen perspective

The planned connection uses your VTA and the Trust Spanning Protocol to carry a scoped exchange for you or your agent. You choose what is presented; the receiving service checks the request before a view is shared.

This guide has no VTA connection adapter yet. Opening Star does not connect an identity or send a key.

Open Star โ†— ยท Inspect your City Key โ†—
guide / Vision / One Logic, Many Skins

One Logic, Many Skins

The same protocol in three registers - engineering, formal, myth - read across one row at a time.

Three bodies of work describe the same thing. Karpathy's The LLM-Wiki and the Self-Improving Research Loop is the engineering register. The agentprivacy distribution gates and trust protocols are the formal register. The The Hitchhiker Expression of the Same Protocol is the mythic register. Below, each concept is read across all three skins.

The correspondence

  • Immutable source of truth -- engineering: raw sources, never edited -- formal: G0, the git canon is upstream -- myth: the recordings, the canon, what was actually said.
  • Selective ingest -- engineering: read a source, file a summary -- formal: G1 classification, circulating vs sealed, manifest-first -- myth: the vision fish translating a signal so it can be heard.
  • Cross-reference integrity -- engineering: maintain links between pages -- formal: G5 hyperlink coherence, every link resolves -- myth: the seam, the trust graph between the cast.
  • Contradiction check -- engineering: lint for stale/conflicting claims -- formal: G2 coherence against the register -- myth: Deep Thought, the long pass that returns a checked answer.
  • Query filed back as source -- engineering: good answers become new pages -- formal: a synthesis re-entered under coherence -- myth: Becoming Real, idea hardening into fact.
  • Gated write-back -- engineering: humans review updates -- formal: G7 fork-back, coherence + attribution + First-Person sign-off -- myth: Don't Panic - a human signs, nothing auto-commits.
  • Credit where due -- engineering: cite sources -- formal: G3 attribution, vendored work credited to its real author -- myth: Who is Involved? the cast named honestly.

Where the protocols sharpen the pattern

The formal register does not just rename the loop - it makes two of its soft parts hard:

  • Coherence becomes cryptographic. A City Key seals a state with a deterministic kappa byte-derivation: the same inputs always produce the same seal, and the seal is checked at every charge. This is lint's "flag contradictions" turned into an equality test a machine can run - coherence you can prove, not just assert.
  • Compounding becomes measurable. The privacy value model V(pi, t) treats value as a moving ceiling that rises as the work accumulates. That is the self-improving loop's monotonic ratchet, given a number. The wiki is not just getting bigger; it is getting more valuable along a tracked curve.

Why hold all three at once

Because each register carries what the others cannot.

  • The engineering skin makes it buildable - any agent with a filesystem can run it tomorrow.
  • The formal skin makes it trustworthy - the gates and the kappa seal mean an unsupervised agent can run 15-file passes without drifting or leaking.
  • The mythic skin makes it adoptable - a community remembers Don't Panic and 42 long after it forgets a gate's number.

The claim of this whole site: an agent plus a persistent, governed wiki is the natural shape of a self-improving research process - and we have now seen it expressed three ways and run on two real corpora. That convergence is the evidence.