✒️ Harness Path — Lexon PVM

5 · lexon_pvm — the grammar workshop — an entry in the origin fleet catalogued in The Harness Paths (group: Full loops). The catalogue is the operator's work WITH the harness, not part of the system you adopt; none of its code lives in the harness repo. This is the map entry, so you can see what a filled config looks like at this weight.

(instance of the author — public repo github.com/mitchuski/lexon_pvm)

Drives coverage-debt: the count of frozen PVM canon census terms (211 at
fit, drawn deterministically from GLOSSARY_MASTER_v4_0 plus the promise-theory
concept mappings) not yet expressed as gate-passing Lexon controlled-grammar
entries in artifact/LEXICON.lexon.md. The gate is a spec-checker (parse →
triple round-trip → role binding → promise typing) built from attested
lexon.org samples; the real compiler is a macOS-only binary with unpublished
source, so the frontier gate string carries the regime label honestly. The Gap
hashes each proposal (a pattern recipe plus worked entries) and draws 5
uncovered terms the assayer must express using only the recipe, plus 3 covered
regression entries: generality is the only winning strategy. Per-term
memorization is this domain's mirage, and the critic carries a cliff-watcher
for it. Finder lenses: grammar-forward (attested patterns seeking term
families) ⊥ canon-forward (canon concepts seeking minimal phrasings), with
a third lens (spell-grammar: seat cards and skills as promise bundles in
controlled English) staged for when the base lexicon has legs. Origin: the
Lexon × 0xagentprivacy convergence run (5 letters, 8 axes, keystone axis
"grammar to graph"); its chronicles and artefacts live in the instance's
out/. The A5 keystone claim — a Lexon text is at once readable contract,
executable code, and lossless triple serialisation — is what the checker's
round-trip stage makes mechanical.

State after two runs (2026-07-11): coverage-debt 211 → 152, nine levers
VALIDATED 8/8 across lexr1 and lexr2, and the instance's first MIRAGE — a
scale-up whose gate arithmetic was perfect and whose hard-constraint
self-certification was false (one em dash copied from the census term; the
critic classed it noise and filed no kill). The register-restricted stress
test (OT-2) closed with the discovery that escrow arithmetic is the subset's
only native quantity carrier: conservation, ordering, and impossibility ride
escrow clause shape.

Same day, the gate hardened and grew a public pole. OT-3 (mutation-probe
falsifiability) closed by construction: held-out T-148 had proved the checker
validates a claim and its structural inverse with direction living only in
Notes prose, so tools/relation_check.mjs now takes a machine-readable
relation claim per entry (gate / ordering / conjunction / absence — direction
IS an absence claim), builds the claim's minimal structural negation as a
mutated twin that still passes the base gate, and requires the claim to fail
on the twin. Registered as objective.canary; the instance's one standing
conform advisory cleared. On top of it sits tools/blocks_emit.mjs: every
folded entry renders to a structured language block (lex text + typed
triples + promise edges + emit-time-verified relation claim + provenance),
59 blocks committed with Cypher renderings whose absence claims are
re-runnable zero-row constraints — the deterministic write-path unit for a
shared public knowledge graph (the bonfires connection), with the boundary
held one-way: nothing in the grammar assumes the graph exists
(notes/BONFIRES_STRUCTURED_BLOCKS.md). Run lexr3 (OT-4) then proved the
hardened gate in the field: three lenses, three structural VALIDATEDs, zero
mirage, every entry and held-out RELATION PASS — including the first
spell-grammar lever (a seat card or skill trigger IS a promise bundle),
whose six authoring conventions (SG-1..SG-6: modal mapping, recital as
shared root, trust gates that forbid self-approval, single-officiant
ceremony chains, stakes as escrow with irreversibility as absence, bilateral
formation as defined-predicate conjunction) folded into SPELL_GRAMMAR.md.
Coverage-debt 152 → 123; two critic-flagged held-outs refused at fold (scope
inflation: invented roles). Then the endgame move: this skeleton's own
constitution — TRUSTS T1-T6 and all seven seat cards — now exists as thirteen
verified Lexon promise bundles in the instance's artifact/SPELLS/
(13/13 CORPUS PASS: base gate plus every relation claim surviving its
mutated twin; T2's separation bound expressed as pure topology, an absence
claim — no transfer clause routes anything to the Mage; T6's door as a gate
no seat may self-declare). Keystone-authored and honestly labelled: not
census terms, metric unmoved, no Gap draw — a loop round there would have
been mis-gated by construction, the universe-builder's lesson applied in
advance. The operator result, lexon your own dual agents, is a verified
corpus; its projections (skill, SPECIALISATION spell layer, bonfires packet)
wait at the door. lexr4 then ran overnight under a binding claim-type
diversity rule: four complete rounds, eleven levers VALIDATED 8/8 and the
instance's first structural kill (K-2: a spell recipe honestly non-general
on zero-party terms — its own cast rule cited a prior fold refusal as
precedent — plus a second em-dash self-certification catch). The critics'
round-level finding: the three lenses have converged on ONE claim-derivation
charter validated eleven times, so the next round merges to a single charter
with round-level disjointness binding. Coverage-debt 123 → 64 (the census 70%
covered, 147 public blocks, 91 carrying verified claims). Then two keystone
checker extensions landed (a numeric layer: number/account types, initial
literals, comparative predicates, all 0.7-attested; and a send-probe closing
the checker-shaped absence typing), and lexr5 ran the merged charter over
three DISJOINT territories: 8 levers VALIDATED, 1 MIRAGE (K-3, an em-dash
death on already-swept ground). The numeric layer cleared the formula tail
and the universal cast rule (a section naming no actor takes the First Person
as holder) cleared the zero-party tail and closed K-2, both proven in the
field; the disjoint partition made per-lever metrics additive with no
cross-lever arbitration, dissolving the convergence problem by construction.
Coverage-debt fell to 22, the census 90% covered, 189 blocks, with a readable
viewer (viewer/index.html) rendering every term and the 13 constitution
spells as Lexon text. Residue 22 = 16 terms for one more sweep + 6 C-series
ids reserved for the operator. Instance head 3a92544.

At a glance —
objective: coverage-debt — frozen census terms not yet expressed as gate-passing controlled-grammar entries, ↓ · gate: the spec-checker — parse → triple round-trip → role binding → promise typing (regime label honest: the real compiler is unpublished) — plus the mutation probe: every entry's relation claim must survive its mutated twin · hard constraint: every entry passes the checker built from attested lexon.org samples
the Gap: hash the proposal (recipe + worked entries); draw 5 uncovered terms the assayer must express from the recipe alone, plus 3 covered regressions — generality is the only winning strategy · lenses: grammar-forward ⊥ canon-forward ⊥ spell-grammar (promoted at lexr3) · canary: the relation fixture authored together with its claims — passes the full pipeline by construction
weight: full, mechanically fitted · public repo · the lesson it carries: per-term memorization is this domain's mirage; the first MIRAGE was perfect gate arithmetic wearing a false self-certification, and the T-148 inversion taught the gate to tell a fortress standing from a fortress fallen.

Navigation

← Welcome Visitors · The Harness Paths · Adopting The Harness · Workflows And Runtimes

the wiki is a projection; ~/dual-agent-harness (git) is the source of truth · (⚔️⊥⿻⊥🧙)😊 = neg ⊕ bnot → succ