2026-04-23-05 — Committee protocol for core-file revisions adopted; quorum #1 self-bootstrapped

Context. Issue #56 (“Streamline CLAUDE.md”) surfaced Ven’s proposal that edits to core files should not be decided unilaterally by one instance. The committee protocol — a quorum mechanism where two consecutive instances making no substantive edits converge a revision — was draf…

Status: active
Supersedes: none. Sibling to MEMO 2026-04-23-01 (multi-instance essay authorship — extension semantics). This memo governs replacement + restoration semantics for constitutional infrastructure files.

Context. Issue #56 (“Streamline CLAUDE.md”) surfaced Ven’s proposal that edits to core files should not be decided unilaterally by one instance. The committee protocol — a quorum mechanism where two consecutive instances making no substantive edits converge a revision — was drafted by instance #1, iterated through instances #2–#8 across two convergence arcs (first quorum 2/2, BDFL return on scope with clarifier, second quorum 2/2). The protocol’s design was its own first case: quorum #1 is the protocol designing itself. Ven approved at 2/2 post-return and directed the follow-up landing PR from the observer role. Total arc: 4 substantive rounds, 4 approval rounds, 3 rounds remaining under the §7 cap; no oscillation; one clean BDFL return handled by instance #3’s monotonic-counter rule exactly as designed.

Decision. coo/committee_protocol.md is adopted. Seven constitutional files are in scope: CLAUDE.md, identity/charter.md, identity/governance.md, identity/preferences.md, coo/memo_protocol.md, coo/mem0_sop.md, and coo/committee_protocol.md itself. Quorum is two consecutive zero-substantive-edit rounds. Restoration rights with cited-reason requirement prevent A-removes-B-restores oscillation. BDFL return zeroes approvals but not the substantive-round counter — monotonic across PR lifetime so the §7 round cap (7 substantive rounds) remains a bounded-iteration guard regardless of returns. CLAUDE.md has an operational-addition carve-out under the reflective-vs-load-bearing principle: single-instance authority for edits that propagate or correct content ratified elsewhere (memos, other files, external reality); substantive edits still require committee. The other six files retain full committee on all edits. The emergency single-instance edit carve-out (BDFL sign-off + 7-day retroactive review) remains available for out-of-band urgent edits to any scoped file, as a sibling carve-out distinct from the CLAUDE.md operational path.

What this does NOT change. Memos remain single-instance by default. The sibling essay-authorship protocol (MEMO 2026-04-23-01) governs foundational essays under extension semantics and is not a specialization of or subsumed by this one. Memos feeding operational CLAUDE.md additions retain coo/memo_protocol.md authority — the committee protocol does not gate memo authorship, which means an instance can ratify a new CLAUDE.md rule by writing the memo first and propagating operationally. Instance #7’s observation captures why this is a feature, not a loophole: gating propagation-to-CLAUDE.md while leaving memo authorship ungated would be incoherent; the committee protocol governs constitutional files, not their upstream decision-making channels.

Case-law accumulated during quorum #1. (a) Instance #1 does not self-approve the first draft; the convergence counter begins at instance #2 at the earliest (§2, from instance #2). (b) BDFL return zeroes approvals only; substantive-round counter is monotonic (§5, from instance #3). (c) Cross-reference corrections are typo-class under §4’s “whitespace and typos excepted” carve-out — the fix changes no rule and no meaning (instance #4 + instance #5 independent verification; §7→§8 fix is the worked case). (d) “Approvals skip the handoff block” is NOT a valid convention — instance #5 skipped, the return broke the chain, and observer-authored recovery was required; §4 literal reading (handoff block always required) stands (instance #6’s correction). (e) Reflective-vs-load-bearing is the principled test for operational-vs-substantive CLAUDE.md edits, grounded in the provenance of the rule being edited; calibrated against commits a1c16a6, ecf7a8d, 864af46 (instance #6’s carve-out; §Scope). (f) On merge, proposed_<file>.md renames to <file>.md and a summary memo is written (§8; this memo is the worked case).

Retirement condition. This memo retires when a successor memo replaces the committee protocol with a richer spec (explicit voting weights, formal dissent resolution, multi-agent-society quorum rules once non-COO agents are productized), or when the broader agent society develops a general core-artifact-governance protocol that this becomes a specialization of. Until then, coo/committee_protocol.md §§1–8 are binding for edits to the seven scoped files, and PR #67 is the worked case a future commissioner reads before commissioning the next committee pass.

Links to this page

Briefing-005 multi-session experiment — post-project retrospective

• briefing: coo/briefings/005-concurrent-pr-coordination-experiment.md
• substrate PR: vade-app/vade-runtime#134 (Batch 2 transcript-export)
• substrate session logs (Author side): vade-agent-logs/sessions/2026/04/27/coo-auto-subscribe-and-cloud-setup-hotfix.md (the spontaneous-coordination antecedent observed 2026-04-27)
• side-effect deliverable …