⭐ Harness Path — Star Hold
17 · star-hold — a coherence instance over a ZK construction (added 2026-09-11) — an entry in the origin fleet catalogued in The Harness Paths (group: Descendant lanes). 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 at ~/dtgwg-zkp-tf-mage/runtimes/star-hold/harness/, over the
runtime beside it; the DTG ZKP task force's lab, the author as co-chair)
The artifact is not prose and not a circuit: it is a claims register over a
reference runtime — nineteen load-bearing claims about a sidecar of signed
envelopes beside a City Key (the "Hold") and its k-of-n community-vouch
presentation, each row naming the H-test that backs it or the condition that
would settle it. The objective is unbacked claims: rows whose enforcement isprompt, manual or nothing — a claim stronger than its enforcement, the
claims checker's own rule made the frontier. Baseline 4 of 19 (the X3 cost of
Ed25519 in-circuit, unlinkability beyond commitment freshness, whether
hidden-value equality is carded as its own primitive, the pairwise-freshness
chokepoint that lives at issuance). The gate is a census: every row probed,
so a claim silently weakened or a test silently renamed cannot hide; the
canary is the runtime's own 45/45. Two lenses: a circuit-mapper backs an
OPEN claim with a runnable check or a measurement; a refuter tries to make
a PROVEN claim's test print FAIL — and a confirmed falsification is a WIN filed
as a K-id, which is the hard constraint's whole point: a claim leaves the
register only by being falsified, never by being relabelled. Conform PASS;
no round has run.
A second instance sits beside it, harness-disclosure/, and it is #11's
seat brought to the Star: hearthold_mage's disclosure-debt rule — canonical
bytes of what a verifier learns, against a frozen requirement census checked
one entry at a time from the public side only — applied to the Star's k-of-n
presentation instead of a Warden's attestation bundle. Baseline 521 bytes
disclosed against 2,529 for the same two credentials shown whole and 6,478
held; a 14-requirement census frozen by sha256; canary 14/14. The seat's own
signed acceptance rides in the Hold as an item the Hold cannot yet verify and
says so.
At a glance —
objective: unbacked claims, lower is better · gate: census overclaims_register.md— PROVEN/DERIVED rows must name H-ids that PASS innode ../test.mjs, OPEN/REPORTED rows must state their settling condition · hard constraint: no claim leaves except by a falsifying K-id; the disclosure set and the rejection register only grow by a live-triggered test
the Gap: the census draw over the ORIGINAL register — the proposer never sees which probe checks which claim · canary: the register +test.mjsas they stand (36/36) · lenses: circuit-mapper · refuter · frontier: 4/19 → OT-1 = a first measurement for the spec's case X3
weight: full loop, unrun · local lab, git-excluded · the lesson it carries: a claims register is an artifact a harness can fold — the metric is honesty about enforcement, and a refuter's win lowers nothing, which is what keeps the register from being gamed by deletion.
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