📉 The Results — Five Runs, 211 To 22

The ladder from 211 to 22, what crossed into constrained language, the two
kills, and the honest residue.

The ladder

The metric is coverage-debt: how many of the 211 frozen canon terms are
not yet expressed as gate-passing Lexon. It started at 211 and only ever
falls when a genuinely general expression survives the full gate.

Five loop runs, twenty-nine validated levers, two structural kills, zero
unverified entries at any fold. Every number resolves to frontier.json.

What crossed into constrained language

The 189 covered terms are, by domain:

• The core architecture — First Person, Sovereignty, the Privacy-Delegation
  Paradox, the Gap, the two agents, the Superagent, the Separation Theorem,
  the Six Dimensions of Sovereignty.
• Promise theory, complete — the whole glossary block (Autonomy Axiom
  through Valency) and all eight promise-theory concept-mapping rows. The
  theory the model rests on is fully expressed.
• The formal core — Entropy, Mutual Information, the Reconstruction
  Ceiling and Error Floor, the Separation Matrix, the three-axis / three-Phi
  separation formalism, Tetrahedral Sovereignty, the Four Forces.
• The consent and standards stack, complete — the entire IEEE 7012 /
  MyTerms cluster (First Party, Agreement, Contract, Entity, Proposer, the
  Agreement Taxonomy, Customer Commons) plus TSP, x402, ERC-8004, ERC-7812.
• The relationship, ceremony, mana, and city layers — VRC, trust tiers,
  the Guardian, the ceremony system, the mana economy, City of Mages
  governance, the V5 value calculus.
• The security and temporal narratives — the Quantum Threshold, the 2D
  Fortress, the 62-Lap Theorem, the Amnesia Protocol, the Temporal Thesis.
• The formula and geometry tail — Conditional Independence, ZKP, the
  holographic and toroidal geometry, BRAID — expressible once a numeric
  layer (number types, initial literals, comparative predicates, all attested
  from Diedrich's own 0.7-era sample) was added to the grammar.
• The cosmological and story terms that name structure the model owns —
  Yggdrasil, the Theia Partition, the thirteen named vertices — expressible via
  a universal cast rule (a canon section naming no actor takes the First
  Person as holder; no role is ever invented).
• The harness's own constitution — TRUSTS T1-T6 and the seven seats, as
  thirteen verified spells.

The two kills, filed at win-prominence

A research instrument earns trust by how it handles failure. Two levers were
killed, each recorded as prominently as any win, each with a re-open
condition:

• K-2 — a recipe for agent-behavior terms that honestly could not express
  a class of "zero-party" canon terms (sections that name no acting party).
  Its own rule forbade inventing a role to fill the gap, and it cited a
  previous run's refusal as binding precedent. It died on that honesty, plus
  one misplaced em dash. Its re-open condition — a "universal cast rule" —
  was exactly what the next run built, and the kill was reopened and resolved.
• K-3 — mechanically perfect (every claim passed), killed on a single em
  dash in a term field. But its ground had already been covered by earlier
  levers in the same run, so nothing was lost. A clean example of the rule: a
  result that violates the hard constraint is not a result at any score.

Two ideas the runs proved out

• A claim can be proven distinguishable from its opposite (the mutation
  probe). This is what makes privacy-as-architecture checkable rather than
  asserted; it is the project's central technical result.
• Convergence is a real risk, and disjoint territories dissolve it. By
  lexr4 the three proposal "lenses" had quietly become one method, so their
  metrics double-counted. lexr5 solved it structurally: it partitioned the
  remaining terms into non-overlapping territories, one per lens, so the
  metric became additive with no arbitration. A methodological lesson with
  general reach.

The residue: 22 terms, and why they remain

• 16 ordinary terms the territory sweeps did not reach in the runs that
  fit before the session limits. One more pass should close most; anything
  that resists a full dry run becomes honest, named, permanent residue.
• 6 conjecture-register terms deliberately reserved. Their canon titles
  carry conjecture numbers and confidence percentages, and no automated seat
  is permitted to write those (the conjecture register is operator-write-only,
  a standing rule). Expressing them needs a human naming convention first.

What "covered" does and does not mean

Two honest caveats, stated plainly:

• Covered means the census, and the census is the canon's index, not its
  body. The 211 terms were frozen from the glossary headings and the
  promise-theory mapping rows. The conjecture register, the full model
  equations, the papers, and the ten Tomes of story are not censused. Coverage
  is over the index.
• Per term, covered means the verified structure, not the substance. Each
  entry carries one machine-checked structural relation and names what it
  omits. The argument of the model is in constrained language; the deep
  mathematical frontier and the myth are, deliberately, left home.

The What This Means For Agentprivacy
develops what this means for agentprivacy; the
The Corpus Coverage Plan lays out the path to fuller
coverage, phase by phase.

Assets

Navigation

← Welcome Visitors · The Grammar Workshop · Reading The Language

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