⚖️ The Grammar Workshop

What if the ideas behind a privacy architecture could be written in a
language that is at once plain English, executable contract, and a
machine-checkable claim? That is what this repository does. It takes the
Privacy Value Model (the theory behind
0xagentprivacy: two separated agents, a Swordsman who protects and a Mage who
delegates, held apart by a Gap) and expresses its canon, term by term, in
Lexon — Henning Diedrich's controlled-English
language for computational law.

The result, after six days of work: 189 of the model's 211 canon terms are
now expressed in a form where the load-bearing claim can be mechanically
verified — and falsified. The census is 90 percent covered. The harness's
own constitution is written in the same language and passes its own gate.

See it, don't just read about it. Open viewer/index.html (viewer/index.html in the repo)
in a browser: every covered term and the thirteen constitution spells,
rendered as Lexon text with its verified claim beside it. Filter by claim
type, search, click to read.

The three ideas, briefly

1. The overlap with Lexon. Every Lexon sentence parses to a
   subject-predicate-object triple — a knowledge-graph edge by another name.
   A contract is a subgraph; its obligations are typed, party-addressed
   edges. So a privacy model written in Lexon is simultaneously readable
   prose, a promise graph, and code. Controlled grammar turns out to be the
   missing authoring layer between how agents talk and how knowledge is
   stored. → The Convergence — Lexon Meets The Privacy Value Model

2. The harness. The work was not done by hand. It ran as a dual-agent
   research loop — the same Swordsman ⊥ Mage architecture the model
   describes, turned on the model itself. One agent proposes ways to express
   canon terms; a third seat (the Gap) hashes each proposal and draws
   held-out terms the prover must express from the recipe alone; a prover
   runs the full gate; a critic names what went wrong. Only genuinely general
   expressions survive. → The Harness — The Model Turned On Itself

3. The result: privacy-as-architecture, made checkable. The model's
   central claim is that privacy is structural — "cannot, by design," not
   "promises not to." In this grammar that becomes a mechanical property: a
   privacy guarantee is a route that provably does not exist, and the tool
   proves it by building the claim's opposite and showing the grammar can
   tell them apart. → The Gate — How A Claim Gets Proven

The ladder (211 → 22)

The remaining 22: sixteen ordinary terms for one more pass, and six
conjecture-register terms deliberately reserved (their titles carry numbers
an automated seat is not allowed to write). → The Results — Five Runs, 211 To 22

Start here

• New to the project? Read docs/index.md — a guided path
  through the whole thing in about fifteen minutes.
• Want to see the language? Open viewer/index.html (viewer/index.html in the repo).
• Want the honest assessment? What This Means For Agentprivacy
  and the Capstone — The Grammar Workshop.

Layout

README.md            this file
docs/                the human-readable wiki (start at docs/index.md)
viewer/index.html    the interactive lexicon reader (self-contained)
artifact/
  CENSUS.json        the frozen 211-term canon census
  LEXICON.lexon.md   the semantic base: every covered term as Lexon text
  SPELLS/            the constitution: TRUSTS + seats as 13 verified spells
  GRAMMAR.ebnf.md    the attested Lexon subset the checker accepts
  OT7_TERRITORIES.json  the residue partition for the merged-charter runs
blocks/              189 portable "structured language blocks" (JSON + Cypher),
                     each a κ-addressed holon (re-derive the address, don't trust the emitter)
tools/
  lexon_check.mjs    the grammar gate (parse, round-trip, roles, promises)
  relation_check.mjs the mutation probe (claims survive their negation)
  blocks_emit.mjs    LEXICON entry -> portable κ-addressed block (holon)
  blocks_audit.mjs   the mesh auditor: re-derive every block's κ (never trust, re-derive)
  kappa.mjs          the κ content-address law (shared with the harness holon layer)
  spells_check.mjs   verify the constitution corpus (13/13)
  render_lexicon.mjs regenerate the viewer
  coverage.mjs       the coverage-debt metric
chronicles/          per-run records + the reflection and capstone
out/                 the original convergence-run artefacts
notes/               the corpus plan, the bonfires write-path spec
frontier.json        the single source of truth for every number

Verify it yourself

node tools/lexon_check.mjs --selftest     # the grammar gate: goldens pass, mutants fail
node tools/relation_check.mjs --selftest  # the mutation probe
node tools/spells_check.mjs               # the constitution: 13/13 CORPUS PASS
node tools/coverage.mjs                   # coverage-debt (22) against frontier.json
node tools/blocks_emit.mjs --selftest     # the public blocks

Gate regime, stated honestly

The checker is a spec-derived validator of an attested subset of Lexon, not
the Lexon compiler (which is a macOS-only binary with unpublished source;
frontier.json carries the regime label). Golden fixtures in tools/golden/
are example texts from lexon.org, credited to Henning Diedrich / Lexon Labs
(and Idelberger for the license example); see SOURCES.md. Because arithmetic
is never executed here, quantitative claims carry a formula's structure, not
its computed values, and every entry names what it omits.

Credits

Lexon is the work of Henning Diedrich and Lexon Labs (lexon.org). The Privacy
Value Model and the Swordsman ⊥ Mage architecture are 0xagentprivacy /
agentprivacy.ai. This repository is the convergence of the two, run through a
dual-agent harness.

Assets

Navigation

← Welcome Visitors · The Convergence — Lexon Meets The Privacy Value Model

the wiki is a projection; github.com/mitchuski/lexon_pvm (git) is the source of truth · a Lexon x Privacy Value Model workshop, run through the The Harness — The Model Turned On Itself