[Governance] Make protected-PR review policy solo-maintainer compatible without weakening deterministic gates

[seonghobae]

Current decision

ContextualWisdomLab is operated by one human maintainer. The current organization ruleset 18156473 applies required_approving_review_count: 1 to ~DEFAULT_BRANCH while required_reviewers is empty, code-owner review is disabled, and no independent eligible human reviewer/team exists. Self-approval is forbidden and model/bot reviews are advisory rather than human approvals.

This is a central governance defect, not a user staffing decision and not a leaf-repository blocker to be worked around. The central .github owner must make the rule structurally satisfiable for a solo-maintainer organization while preserving or strengthening deterministic evidence gates.

Required causal repair

1. Re-read the live organization ruleset and affected repositories before mutation; do not rely on remembered review counts or branch names.
2. If no real eligible independent human authority exists, remove or replace only the structurally impossible generic approving-review-count requirement for the affected solo-maintainer repositories/default branches.
3. Keep self-approval forbidden. Do not represent CodeRabbit, OpenCode, Noema, GHAS, Dependabot, or another model/bot as a human approval.
4. Preserve required workflow/security/SAST/coverage/package/SBOM/provenance/exact-head binding, unresolved-thread resolution, deletion protection, non-fast-forward protection, and routine bypass prohibition.
5. Keep stale/predecessor/queued/skipped/cancelled/absent/model-only evidence non-passing.
6. Use guarded administrator bypass only for a separately proven chicken-and-egg control-plane repair that meets the organization emergency policy; ordinary product PRs and ordinary approval waits are not bypass cases.
7. After the ruleset mutation, re-read the effective policy from each affected live default branch and prove an unchanged exact-head PR can become mergeable through deterministic gates alone without self-approval or fabricated reviewers.
8. Record the before/after ruleset JSON, affected repository/default-branch identities, exact validation PR/head, required check set, and resulting protected merge evidence.

RED acceptance

A current protected repository owned solely by the maintainer has an unchanged exact-head PR for which all applicable deterministic repository/security gates and review threads are clean, yet merge remains impossible solely because the live ruleset demands one approving review and there is no eligible independent human authority.

Wardnet currently provides this class of evidence: live default/protected branch is main, organization ruleset 18156473 targets ~DEFAULT_BRANCH, and the rule still reports required_approving_review_count: 1, required_reviewers: [], require_code_owner_review: false, require_last_push_approval: false.

GREEN acceptance

• The impossible approval-count deadlock is removed or replaced at the central governance layer.
• Deterministic CI/security/SAST/coverage/package/SBOM/provenance/thread/branch-integrity gates remain enforced.
• Self-approval remains impossible/unused and bots/models are not counted as humans.
• An unchanged exact-head validation PR becomes mergeable only after all then-live deterministic gates are terminal successful.
• The same policy is documented as the current solo-maintainer governance model and no owner automation asks the user to provision fictional reviewers.

Non-goals

• Weakening deterministic security or quality gates.
• Enabling routine administrator bypass.
• Fabricating a second identity or reviewer team.
• Treating model review as human approval.
• Changing product repository source code to work around a central governance defect.

[seonghobae]

Additional protected-branch evidence — mightyETL, 2026-08-06

ContextualWisdomLab/mightyETL#121 is another reproducible instance of the counted-review deadlock described here.

• Exact head: ed88fc8d97bb46dca7cda4c947532b688dcda08a
• Mergeability: GitHub reports mergeable
• Exact-head repository gates: Ubuntu/macOS/Windows CI, Dependency Review, CycloneDX SBOM, SAST Semgrep, Security Scan, and CodeRabbit status all succeed
• Review state: zero unresolved current threads and zero requested-change reviews
• Automation hardening: the OpenCode model job has pull-requests: read; a non-checkout deterministic publisher is the sole pull-requests: write holder; a second non-checkout job is the sole actions: write holder; workflow-run authorization binds both exact PR number and exact SHA
• Review request path: Draft → Ready was retriggered and exact-head requests were posted to OpenCode and Noema
• Remaining blocker: no non-author APPROVED review that GitHub counts is anchored to the exact head

The downstream dependency chain is also prepared and green on its stacked targets:

1. ContextualWisdomLab/mightyETL#122 — f57256bcfb89c443415709a4af30ce7b190684e1
2. ContextualWisdomLab/mightyETL#129 — 349afe7f0e9c77279806f13b2d156e4bcd43bab1
3. ContextualWisdomLab/mightyETL#130 — 1cd1dbda288f57906ef00669f5de78393a29e16f
4. ContextualWisdomLab/mightyETL#131 — aa7bb331faf8e7b3be439a17c36bafed9fc9638d

Each passes cross-platform CI, Dependency Review, and SBOM on its current stacked base. CodeRabbit completed exact-head static reviews of ContextualWisdomLab/mightyETL#122, ContextualWisdomLab/mightyETL#129, ContextualWisdomLab/mightyETL#130, and ContextualWisdomLab/mightyETL#131 with no actionable or blocking findings. They remain intentionally unmerged and will require direct-develop SAST/Security reruns and fresh counted approvals after predecessor integration.

This evidence supports the issue's proposed remedy: provision at least two genuinely independent human reviewer identities with the minimum counted permission and team-based rotation. Do not solve this by self-approval, administrator bypass, broad bot contents-write access, a second credential representing the author, or weakening required review count.

[seonghobae]

Current blocking evidence — mightyETL, 2026-08-07

The complete linear stack is technically green on every check that materializes for its current target, but the root remains unable to satisfy the counted-review rule:

1. mightyETL#121 — 2f374446b0e0dc180c53736787a2a7a9b331503f → develop
2. mightyETL#122 — 2516c3f945df8fee7ad451319a20b63dd28fb3d8
3. mightyETL#129 — 683e1a8b132772236b561b28d5d204c866a08034
4. mightyETL#130 — e7495b4fe558f543fe5fc1a1834e2f87542263a1
5. mightyETL#131 — addfe57e8948a8e274b297cf629582895fb760fc
6. mightyETL#133 — ee50859d654b89fdba5b14424fe7b8afc3b6c99d
7. mightyETL#135 — 839fa54cdc0c302c053842b26f3c36a16764d8ae

For root PR #121, exact-head CI on Ubuntu/macOS/Windows, Dependency Review, CycloneDX SBOM, SAST Semgrep, Security Scan, and CodeRabbit status are successful. GitHub reports it mergeable, all visible inline review threads are resolved, and no requested-change review remains. The formal review inventory still contains no qualifying non-author APPROVED review.

I attempted the available reviewer-assignment paths without changing code, permissions, or protection rules:

• individual requests to cwl-noema-review and opencode-agent returned GitHub 422: they are not repository collaborators;
• a team request to independent-reviewers returned GitHub 422: that team is not a collaborator on ContextualWisdomLab/mightyETL.

The six downstream exact heads each have successful current stacked-target CI, Dependency Review, CycloneDX SBOM, CodeRabbit status, mergeability, and zero unresolved threads, but no formal approval. Their non-default bases intentionally do not materialize Semgrep/Security Scan; those gates must rerun after each predecessor merges and the next PR is retargeted to develop.

No approval was synthesized, no author identity was reused, no manual-merge hold was removed, and no branch/ruleset bypass was attempted. The actionable organization-level prerequisite is therefore exactly this issue’s acceptance criterion: provision a genuine non-author human reviewer team with counted repository permission, then request an exact-head review of #121. Once #121 merges, the remaining stack can be retargeted, fully revalidated, independently approved, and merged in order.

[seonghobae]

Additional blocked stack — mightyETL

ContextualWisdomLab/mightyETL#121 remains the dependency-eligible root of a seven-PR durable-job stack. Its exact head recorded by the repository is 2f374446b0e0dc180c53736787a2a7a9b331503f; current technical evidence includes successful multi-platform CI, Dependency Review, CycloneDX SBOM, SAST Semgrep, Security Scan, CodeRabbit status, mergeability, and zero unresolved review threads. The remaining merge gate is a GitHub-counted non-author APPROVED review anchored to that unchanged SHA.

The unavailable counted-review path blocks #122, #129, #130, #131, #133, and the Draft replay PR #135 in dependency order. No bot, second credential for the author, administrator bypass, reduced review count, stale approval, or predecessor-head evidence is an acceptable substitute.

Please include mightyETL in the independent-reviewer-team dry run and evidence record: reviewer identity and current counted permission, exact head SHA, required-check set, review timestamp, merge SHA, stale-review invalidation after a later commit, and fail-closed behavior when reviewer permission is removed. The existing OpenCode, Noema, CodeRabbit, GHAS, and repository-specific review credentials must remain advisory and unchanged.

[seonghobae]

Current central bootstrap evidence — .github#823

The counted-review deadlock now reproduces on the organization control-plane repository itself.

• Pull request: ContextualWisdomLab/.github#823
• Exact unchanged head: a3876dffb3669e6187d5c20af6ecf64f41faf11a
• Protected main: 1131b1bbafb24e455fc8619cdf316813e8721861
• GitHub reports the PR open, non-draft, and mergeable.
• Exact-head Strix Changed Path Quality CI, CodeQL, Python Security, Security Scan, Semgrep, Secret Scan, OSV, Scorecard, and SBOM evidence is terminal successful; the current-head OpenCode review reports APPROVED, but it is advisory and does not replace the required counted independent approval.
• A direct review request to team independent-reviewers was attempted on the unchanged exact head and GitHub returned HTTP 422: Reviews may only be requested from collaborators. One or more of the teams you specified is not a collaborator of the ContextualWisdomLab/.github repository.

No approval was synthesized, no author identity was reused, and no branch/ruleset bypass was attempted. The actionable organization prerequisite is unchanged but is now confirmed for the central source repository as well: grant a genuinely independent human reviewer team the minimum permission GitHub counts on ContextualWisdomLab/.github (and the other protected CWL repositories), then request a fresh exact-head review. This is especially blocking because #823 repairs the required Strix quality timeout budget that currently leaves multiple downstream central PRs cancelled despite otherwise successful exact-head gates.

[seonghobae]

Current governance reproduction on .github#790: exact head 28a878f205002ba458c1be476353a3680637cf69 is Ready/mergeable, protected main 6eb06cdd08c79a06f7b390069d4ffa49e2eb7dba is an ancestor, exact-head Trusted uv Materializer Quality CI and Strix Changed Path Quality CI are terminal-success, all current security/supply-chain workflows are terminal-success, CodeRabbit status is success, and review threads are resolved. No qualifying independent non-author formal APPROVED review is present. A fresh review request to team independent-reviewers returned GitHub HTTP 422: Reviews may only be requested from collaborators because the team is not a collaborator of ContextualWisdomLab/.github. This reproduces the issue #772 governance deadlock without changing branch protection or reviewer credentials. Required external action remains granting a real independent human reviewer/team the minimum permission GitHub counts for protected approval; do not substitute author self-review, advisory bots, or bypass.

[seonghobae]

Fresh feasibility evidence (2026-08-08): repository collaborator-permission lookup on ContextualWisdomLab/.github reports opencode-agent=none, coderabbitai=none, and no standard collaborator permission for cwl-noema-review. Protected-main .github/CODEOWNERS is still owned only by @seonghobae. This confirms the current deadlock is governance/authority, not a missing PR-side repair. Granting advisory bots broad write merely so GitHub counts their approvals would expand autonomous authority and defeat the intended separation of duties, so that candidate is rejected. The realistic durable path remains at least two independent human reviewers (team/custom role with the minimum permission GitHub counts), exact-head/stale-review enforcement, rotation/offboarding, and a dry-run proving author/advisory-bot approval does not count while a genuinely independent reviewer does. Until that governance prerequisite exists, PR-side automation must continue to treat bot reviews as advisory evidence only and must not bypass the counted approval rule.

[seonghobae]

Fresh counted-approval deadlock evidence — central PR #834

Revalidated on exact current head 7e3c9e581f981bd2ceaa37c7b9bde16b452b899b over protected main 6eb06cdd08c79a06f7b390069d4ffa49e2eb7dba.

RCA: #834's repository-side Noema OIDC repair is no longer blocked by a source, test, security, supply-chain, or unresolved-review defect. After the latest test-contract hardening, exact-head Scorecard, SBOM, OSV, Secret Scan, Security Scan, Semgrep, CodeQL, Python Security, and Strix Changed Path Quality CI all completed successfully. CodeRabbit's current-head incremental review reports no actionable comments, its exact-head commit status is success, and both prior actionable inline threads are resolved/outdated. GitHub reports the PR Ready and mergeable.

Approval feasibility was rechecked rather than inferred:

• protected-main .github/CODEOWNERS still assigns *, .github/workflows/*, and scripts/ci/* only to @seonghobae;
• the PR initially had no requested reviewer or requested team;
• a fresh request for team independent-reviewers on this exact head returned GitHub HTTP 422: Reviews may only be requested from collaborators; the team is not currently a collaborator of ContextualWisdomLab/.github;
• the formal review inventory contains no qualifying non-author APPROVED review on the exact current head;
• opencode-agent collaborator permission is none;
• coderabbitai collaborator permission is none;
• cwl-noema-review has no returned counted repository permission.

Rejected remedies: author self-approval, a second author credential, treating COMMENTED/status/model output as approval, granting a bot broad write solely to manufacture approval, reducing the review gate, or bypassing branch protection. None preserves separation of duties.

The smallest realistic remaining prerequisite is therefore external organization governance: provision a genuine non-author human reviewer/team with the minimum permission GitHub counts, add that team as a collaborator on this repository, then obtain an exact-current-head approval. After that approval, normal protected merge may proceed; #834 still requires its separately defined protected-main real OIDC exchange as post-merge operational acceptance. No bot review is being promoted to the counted gate.

[seonghobae]

Fresh governance evidence at protected main 6eb06cdd08c79a06f7b390069d4ffa49e2eb7dba (2026-08-12):

• GitHub's current indexed open-PR search returned 47 PRs; 28 are Ready and mergeable.
• A separate current review:approved search returned zero open PRs.
• Protected-main .github/CODEOWNERS assigns *, workflows, and scripts/ci/* only to @seonghobae, who is also the author of the blocked PRs.
• Automated CodeRabbit/OpenCode/Noema evidence remains advisory and cannot satisfy the required independent human approval.

This re-proves the first failing boundary as reviewer eligibility/permission, not source, CI, or mergeability. The smallest safe remedy still requires organization authority: create and permission a real non-author human review team (at least two active people), then prove counted exact-head approval on a dry-run PR. No bot authority, approval count, protection, or bypass setting was changed.

[seonghobae]

Additional counted-review deadlock evidence — OriginWeave, 2026-08-16

Fresh live repository evidence reproduces this governance deadlock in ContextualWisdomLab/OriginWeave without a source/test/security blocker.

Ready technical roots

• OriginWeave#37 exact head fe324f71ccd154f7759f009d8c34fb6247c7d070 targets protected main 0c376acf059be9ddddddfbde1d0189e4f39ef014, is mergeable, and is ahead 96 / behind 0. Its current-head Rust contracts, production coverage, central coverage-evidence, OpenCode, Noema, Strix, SAST/Semgrep, Trivy, OSV, Scorecard and dependency-review evidence are successful; current inline review findings are resolved. No counted independent/latest-push APPROVED review is present.
• OriginWeave#40 exact head 90227c5dd6c62839d4ccfe86640f08984011e207 targets the same protected main, is mergeable, and is ahead 38 / behind 0. Exact-head CI, SAST/Semgrep, Security Scan, Manifest V3 compatibility, OpenCode and coverage-evidence are successful; current inline review findings are resolved. No counted independent/latest-push APPROVED review is present.

Live reviewer authority

A fresh collaborator inventory returns only the author/administrator identity seonghobae; the direct-collaborator inventory is empty. The live ruleset CWL Central required workflows requires one approving review, dismisses stale approvals, and requires approval after the latest push, with no bypass actor configured. Protected-main AGENTS.md also forbids scheduled agents from manufacturing/self-approving or weakening the gate.

This is therefore another concrete acceptance case for #772: provision a genuinely independent non-author human reviewer path with the minimum permission GitHub counts, then obtain an approval anchored to each unchanged exact head. Do not solve it by administrator bypass, author identity reuse, bot privilege expansion, or reducing the required review count.