[Review] Obtain two independent M0 protocol and safety reviews

[veil-chow-fyaic]

Goal

Satisfy the M0 exit criterion requiring two independent design reviews before declaring the foundation implementable without undocumented assumptions.

Reviewer brief

Review the protocol from two perspectives:

1. distributed-systems correctness: idempotency, ordering, expiry, replay, freshness, lifecycle, and audit;
2. agent/harness safety: context sovereignty, provenance, consent, structured gates, wake/interrupt budgets, and cross-harness degradation.

Evidence required

• Reviewer affiliation or relevant perspective is disclosed.
• Findings identify exact documents or schema paths.
• Each finding is accepted, resolved, deferred with rationale, or rejected with rationale.
• At least one review comes from outside the maintainer organization.
• Final review artifacts are linked from the M0 milestone.

[veil-chow-fyaic]

Internal review round — 2026-08-20

Three independent read-only Codex sub-agent review lanes examined main@c38873026222175433fb86f2fcac1a655ffcc932:

• distributed-systems and protocol correctness: request changes;
• authorization, provenance, privacy, and consent safety: request changes;
• cross-harness implementability and Kimi feasibility: request changes.

The reports are being added under docs/09-reviews in the current implementation PR. They found no P0 issue, but identified M0 blockers covering crash/idempotency semantics, authenticated principals and grants, revocation-at-admission, cross-state CAS, mailbox/registration bindings, partial interruption, external verification, summary disclosure, and capability coherence.

These internal passes do not satisfy this issue’s organizationally independent external-review criterion. Kimi Code was also attempted as a different-model review/test path: ACP v1 handshake against local Kimi Code 0.36.1 passed, but the live model call returned HTTP 403 because the account billing-cycle quota is exhausted. This issue remains open.

[veil-chow-fyaic]

External review request is ready

All normative M0 blockers (#15–#19) are now closed. The review target is main@265e461e4f8d53447d21cfe79894ddd2f46d297b; the later docs-only commit publishes the packet.

• Reviewer packet: https://github.com/fyaic/threadmesh/blob/main/docs/09-reviews/m0-external-reviewer-packet.md
• Submission template: https://github.com/fyaic/threadmesh/blob/main/docs/09-reviews/external-review-template.md
• One-command boundary: npm ci && npm test && npm audit --audit-level=high

We are requesting two qualifying verdicts:

1. distributed-systems correctness (idempotency, crash windows, CAS, replay, expiry, lifecycle);
2. agent/harness safety (authority, provenance, context sovereignty, partial cancellation, verifier trust).

At least one reviewer must be outside the fyaic organization. Please disclose affiliation/perspective and relationship to maintainers. A review may be posted here or added under docs/09-reviews/ by pull request. Every finding will receive a public disposition; internal agent reviews are not counted toward this gate.

[veil-chow-fyaic]

Correction: exact review target

The prior request comment contained a one-character transcription error in the commit SHA. The nonexistent value 265e461e4f8d53447d21cfe79894ddd2f46d297b must not be used.

The actual, fetchable merge commit for #25 and the authoritative M0 review target is:

265e461f1b8714c56f7fe817795b81d895f732c6

PR #46 corrected both the reviewer-packet heading and git checkout command on main. Verification: 265e461

No protocol content or review boundary changed; this correction makes the already intended target reproducible. Please use the full corrected SHA in every verdict.

[veil-chow-fyaic]

Gate and discoverability update

The public review request is now visible at the top of both README entry points via merged PR #47. The packet target typo was corrected transparently via #46 and the authoritative target is 265e461f1b8714c56f7fe817795b81d895f732c6.

Draft #45 now contains a machine-checkable review gate. Before any live agent product can start, it requires two integrity-bound public review records with distinct reviewers, both required perspectives, at least one outside reviewer, the exact target commit, approving verdicts, and terminal public dispositions for every finding. An operator environment acknowledgement cannot bypass missing records.

Current truthful gate output remains reviewCount: 0, externalReviewerCount: 0; this issue is still the critical-path blocker.

[veil-chow-fyaic]

Critical-path status — 2026-08-21

All internal implementation/review work needed to consume external verdicts is now complete:

• three independent internal lanes approved the gated product runner and bounded evidence path;
• the proposal→grant transaction finding was moved into feat: add fail-closed relationship policy #30 at reviewed 61c15ae and propagated through the entire feat: add crash-safe durable dispatcher #31–feat: generalize multi-product context admission #43 stack;
• all ten stacked PRs are CLEAN with green CI;
• the checked-in verifier still truthfully reports reviewCount: 0, externalReviewerCount: 0, exit 3;
• no real model validation was started.

The authoritative normative review target remains 265e461f1b8714c56f7fe817795b81d895f732c6, not the later implementation stack. The packet and canonical machine-block template linked above remain current.

This issue is now the only critical-path gate before sequential stack merge/revalidation and the bounded Codex/Kimi/Gemini real-product runs. Internal Codex reviewers are deliberately not counted.

[veil-chow-fyaic]

Maintainer decision: bounded real-product experiments may proceed before this external-review gate closes. The override is explicitly non-normative and does not count the maintainer or internal Codex reviewers as external reviewers. Results carry authorization.mode=maintainer-experimental, normativeReviewSatisfied=false, and an unsatisfied M0 gate projection.

Issue #7 remains open with 0/2 qualifying reviews. This separates product learning velocity from normative review truthfulness.