Witness seat — notes

The third seat on the dream cycle: a VWC (Verifiable Witness Credential) attesting the
collision, transcript-bound, optional, and never a substitute for consent.

Source design: VWC — Witness Seat. This build
closes the oldest open thread in the lab — runtime 07's "next steps" item 1 / candidate-note 3
("a VWC slots straight in"), flagged in the 2026-07-16 dream-cycle chronicle.

What it demonstrates

Runtime 07 grows the trust graph in two seats: 🧙 the Mage proposes the smallest edge, ⚔️ the
Swordsman proves it across the Gap. This runtime seats the third role the cred-spec provides
but the cycle didn't hold: a witness — a registry-anchored third party who sees THAT a
meeting happened, never what it meant. The witness receives only the collision's public
commitment (the candidate's VRC commitment), the two fresh R-DIDs, and the §15.2 canonical
transcript; it signs an attestation binding those four together in its own
personhood-anchored name.

The seat upgrades edge evidence without touching edge admissibility:

• a witnessed edge carries independent corroboration of the collision, re-verifiable later
  from the stored edge + the retained transcript + the witness registry alone (W1);
• an unwitnessed edge forms exactly as rt07 forms it today — the unwitnessed path is
  byte-for-byte rt07's path (W2);
• W4 is the separation property: the attestation is never consulted before, or instead
  of, the Swordsman's verdict. A witness holding a perfectly valid attestation cannot move an
  unconsented candidate past the consent gate, and cannot forge a proposal (it never holds
  the parties' secrets, so it cannot mint their fresh per-counterparty R-DIDs). Witness ≠
  consent ≠ proposal — the seat must not collapse into a third signing authority. This is the
  same separation discipline X9's guardian doc states one level up (guardians attest
  recovery, they do not authorize identity), inherited here at the edge-formation level.

Boundary notes (recap from the design doc)

Assurance boundary — a verifier of a witnessed edge may rely on: a registry-anchored
third party signed a statement binding this collision commitment to this transcript. It
must NOT infer: that the collision's content was as claimed (the witness saw an encounter,
not the shared secret); that the witness was honest; that a witnessed edge is more
consensual than an unwitnessed one. Accountability: the witness for its attestation; the
context authority for who may witness.

Disclosure boundary — the witness is a §19 observer (the observer list gains a row):
it learns that two R-DIDs collided at a time, under a context. It must not learn the shared
secret or either party's M-DID — enforced structurally here (W5: view whitelist + canonical
transcript schema + lifetime-view scan). Witness patterns (who witnesses whom, how often)
are themselves correlatable — a busy witness becomes a traffic-analysis hub; retention rules
for witness logs belong in the profile (§20 minimisation applies).

Named failure codes (fixture-register v3 candidates)

Checked in this order; append-only candidates for ../fixtures/reasons.mjs v3:

Design note: an invalid attestation rejects the witnessed-edge operation by name (an
active forgery is a mirage, like a forged VRC commitment — the Swordsman's posture), while an
absent attestation changes nothing (W2). Absence ≠ forgery.

The additive refinement

The exploration's M1 wording ("implement the witness channel in rt 07") is refined to
own-directory-importing: runtime 07 is a byte-stability dependency of the fixtures suite
(its rejection strings are pinned in the reasons register, byte-for-byte), so this build
lives in Lab — witness-seat, imports rt07's joinCommunity / Swordsman / Mage /
encounter and calls Swordsman.prove UNCHANGED — the lab's established precedent
(show-composition, quiet-presentation, guardian-recovery all wrap rather than edit). W7 makes
the byte-stability contract executable: rt07's own test.mjs runs in a subprocess and must
still report 11/11.

Other deliberate model choices:

• Signature model: H-based MAC-style signature (H(domain, key, canonicalize(body))),
  same modelling honesty as rt07's roster holding node secrets — the witness registry holds
  the verification key; the circuit swaps a real signature scheme, the binding argument is
  identical.
• Collision commitment = the candidate's public VRC commitment (claimedVrc), which the
  wrapper cross-checks against the Swordsman's independently recomputed vrc — the witness
  binds to the recomputed value, never to the Mage's bare claim.
• Transcript = the shared canonical/canonical.mjs §15.2 implementation (the design
  doc's M2: one mechanism, shared with the freshness harness — a bare nonce is insufficient).

Deferred (parked per the design doc)

• Same-VTC-witness question: must a witness hold a VMC in the same VTC as the parties,
  or is any registry-anchored witness acceptable? Same-community witnessing is stronger for
  the community-anchored construction but shrinks the witness pool and increases
  intra-community traffic analysis. (This build anchors the witness in a community via
  rt07's joinCommunity but does not require it to be the parties' community.)
• Weighted propagation: does a witnessed edge earn different weight? Weight semantics
  belong to trust-task protocols, not the formation layer — parked, noted.

Pointers

• Design doc: VWC — Witness Seat (M1–M4; this
  build = M1 + M2, boundary recap above = M3 draft carried; M4 = the upstream candidate-note
  feed, now backed by a runnable reference + W4 stated as a property).
• Thread closed: rt07 NOTES.md "next steps" item 1 + candidate-note 3, flagged in the
  2026-07-16 dream-cycle chronicle.
• Decision doc anchors: §15/§15.1/§15.2 (canonical transcript, replay), §17.2/§7.3
  (separation discipline — attestation ≠ authority), §19 (observer analysis), §20
  (minimisation of witness logs).
• Tests: node test.mjs → 15/15 (W1–W7; W7 = rt07 regression subprocess).