fix(tier4): accept trusted admin collaborators

[scarmani]

Summary

• Allow explicit trusted Tier-4 operator comments from GitHub COLLABORATOR authors when live repo admin permission is verified.
• Keep the trust boundary narrow: OWNER remains accepted, MEMBER keeps admin-permission behavior, non-allowlisted collaborators and non-admin collaborators remain rejected.
• Add regression tests for the org-owned repo case that blocked fix(settlement): route Tier-4 read probes through app auth #8405 despite green quorum.

Validation

• python3 -m pytest tests/scripts/test_settle_tier4_pr.py -> 71 passed
• pre-commit run --files scripts/settle_tier4_pr.py tests/scripts/test_settle_tier4_pr.py -> passed
• bash scripts/automation_pr_preflight.sh origin/main HEAD -> passed
• push hook mypy/pre-commit -> passed

Notes

• Does not change merge policy, branch protection, workflows, settle_one_pr.py, or review queue.
• No merge or admin action performed for fix(settlement): route Tier-4 read probes through app auth #8405.

[scarmani]

Grok independent model review

Reviewer: grok (xai) — independent adversarial model review via the Aragora Grok reviewer, grounded on the exact PR head.
Head: f175778 (f175778), committed 2026-06-14T07:36:44Z.
PR: #8412.
Model family: grok

Verdict: PASS

• No blocking issues found. Logic correctly gates COLLABORATORs behind explicit allowlist plus admin permission check via the collaborators endpoint; message/diagnostic updates and new tests are consistent; no correctness, authorization, or regression risks identified in the diff.

dogfood: yes

[scarmani]

OpenAI/Codex independent model review

Reviewer: codex (openai) — independent adversarial model review via Codex CLI, grounded on the exact PR head.
Head: f175778 (f175778), committed 2026-06-14T07:36:44Z.
PR: #8412.
Model family: openai

Verdict: PASS

• No correctness, authorization, or security findings. The diff keeps COLLABORATOR authorization fail-closed: explicit trusted-operator allowlist plus live admin permission are both required.
• Existing OWNER and MEMBER behavior remains intact.
• Test coverage is adequate for this change: success, missing allowlist, missing admin permission, and REST permission endpoint encoding are covered.

dogfood: yes
Receipt artifact: /tmp/pr8412-codex-direct-review.txt

[github-actions]

Aragora Code Review

Advisory-only review. No issues found.

[scarmani]

Grok independent model review

Reviewer: grok (xai) — independent adversarial model review via the Aragora Grok reviewer, grounded on the exact PR head.
Head: 53314e3 (53314e3), committed 2026-06-14T09:32:30Z.
PR: #8412.
Model family: grok

Verdict: PASS

• No blocking correctness, security, or regression issues found in the diff.
• Logic correctly fail-closed for COLLABORATOR: explicit allowlist plus admin permission check, while preserving prior MEMBER behavior.
• Permission-check path and diagnostics are updated consistently; new tests cover allowlist/permission cases and the REST endpoint.

dogfood: yes

[scarmani]

OpenAI/Codex independent model review

Reviewer: codex (openai) — independent adversarial model review via Codex CLI, grounded on the exact PR head.
Head: 53314e3 (53314e3), committed 2026-06-14T09:32:30Z.
PR: #8412.
Model family: openai

Verdict: PASS

• No blocking correctness, authorization/security, or test adequacy findings.
• The COLLABORATOR path is fail-closed: it only counts with an explicit trusted-operator allowlist entry plus a live admin permission check.
• Unallowlisted collaborators and allowlisted non-admins reject; prior OWNER and MEMBER behavior remains intact.

dogfood: yes
Receipt artifact: /tmp/pr8412-codex-direct-review-53314.txt

[scarmani]

Grok independent model review

Model family: grok
Provider/model: xAI grok-4-latest
PR: #8412
Exact head: 75f9baf478c0e155a813bcd679248285c3934b81
Dogfood: yes - reviewed the current PR diff after the TW03 timestamp repair and verified the exact-head helper/docs changes.
Receipt artifact: /tmp/pr8412-grok-response-75f9.json

Verdict: PASS

Grok verified that the prior TW03 docs-site timestamp issue is resolved: docs-site/docs/contributing/tw03-rescue-productization-status.md remains at 2026-06-04T13:31:36Z, matching docs/status/generated/rescue_productization/latest.json generated_at, and the current PR diff no longer changes that file.

[scarmani]

OpenAI/Codex independent model review

Model family: openai
Provider/model: OpenAI Codex gpt-5.5
PR: #8412
Exact head: 75f9baf478c0e155a813bcd679248285c3934b81
Dogfood: yes - reviewed the current PR diff after the TW03 timestamp repair and checked the Tier-4 helper logic plus publication surfaces.
Receipt artifact: /tmp/pr8412-codex-review-75f9-fast.txt

Verdict: PASS

OpenAI/Codex reviewed exact head 75f9baf478c0e155a813bcd679248285c3934b81 against origin/main, found no blocking issues, and verified the prior TW03 docs-site dissent is resolved: the docs-site TW03 Last updated value is 2026-06-04T13:31:36Z, matching docs/status/generated/rescue_productization/latest.json generated_at, and the PR diff no longer changes that file.

[scarmani]

PR #8412 exact-head model evidence

Model family: openai

PR: #8412
Exact head: ccecd65
Dogfood: yes. OpenAI/Codex verified the HEAD SHA, branch diff versus origin/main, TW03 mirror/source timestamp by shell parse, B0/METRICS source-backed values, and the Tier-4 trusted-admin helper logic. No long suites were run.
Receipt artifact: /tmp/pr8412-codex-review-ccecd65fd2.txt

Verdict: PASS

Blocking findings: none.

Reviewer summary: TW03 mismatch is resolved; docs-site/docs/contributing/tw03-rescue-productization-status.md is not in the PR diff versus origin/main, and at HEAD its Last updated equals docs/status/generated/rescue_productization/latest.json generated_at (2026-06-04T13:31:36Z). Trusted COLLABORATOR settlement remains fail-closed via explicit trusted-operator allowlist plus live admin permission.

[scarmani]

PR #8412 exact-head model evidence

Model family: grok

PR: #8412
Exact head: ccecd65
Dogfood: yes. Reviewed the current-head diff and verified the prior TW03 docs-site timestamp mismatch is absent from the PR diff; at HEAD the TW03 docs-site Last updated value matches docs/status/generated/rescue_productization/latest.json generated_at (2026-06-04T13:31:36Z).
Receipt artifact: /tmp/pr8412-grok-review-ccecd65fd2.txt

Verdict: PASS

Blocking findings: none.

Reviewer summary: the Tier-4 trusted-admin helper change remains fail-closed because COLLABORATOR requires explicit trusted-operator allowlist membership plus a live repo-admin permission check; no correctness/security regression was found.

[scarmani]

Grok independent model review on head 67d674c

Reviewer harness: grok
Model family: grok
Model id: grok-4-latest
Receipt artifact: /tmp/pr8412-grok-review-67d674c749.txt

Independent model review (adversarial dogfood recheck) of PR #8412 at exact head 67d674c.

• VERDICT: PASS.
• Verified Tier-4 COLLABORATOR allowlist plus live admin permission paths, _operator_author_rejection_reason, _trusted_member_requires_permission_check, updated tests, regenerated docs/METRICS.md, and TW03 timestamp alignment.
• No blocking findings in the trusted-admin helper logic, diagnostics, tests, or source-backed docs repair.

[scarmani]

Codex independent model review on head 67d674c

Reviewer harness: codex
Model family: openai
Model id: gpt-4.1-codex
Receipt artifact: /tmp/pr8412-codex-review-67d674c749.txt

Independent model review (adversarial dogfood recheck) of PR #8412 at exact head 67d674c.

• VERDICT: PASS.
• Verified exact HEAD, clean PR diff, git diff --check, scripts/regenerate_metrics.py --check, TW03 docs-site timestamp equality with latest.json, AST parse for helper/test files, and direct collaborator authorization dogfood.
• No blocking findings. Tier-4 collaborator handling remains fail-closed: COLLABORATOR requires explicit trusted-operator allowlist plus live admin permission, and unallowlisted collaborator comments reject.

[scarmani]

Grok independent model review

Reviewer: grok (xai) — independent adversarial model review via the Aragora Grok reviewer, grounded on the exact PR head.
Head: 4850d5d (4850d5d), committed 2026-06-14T23:26:08Z.
PR: #8412.
Model family: grok

Verdict: PASS

• No blocking issues identified. Logic correctly fail-closes on COLLABORATOR (require allowlist + admin perm check via existing REST path); MEMBER behavior unchanged. Tests cover new paths + edge cases; error strings and METRICS.md updates are consistent.

dogfood: yes

[scarmani]

OpenAI independent model review

Reviewer: openai (openai) — independent adversarial model review via Codex CLI OpenAI harness, grounded on the exact PR head.
Head: 4850d5d (4850d5d), committed 2026-06-14T23:26:08Z.
PR: #8412.
Model family: openai

Verdict: PASS

• No blocking issues found.
• [P3] scripts/settle_tier4_pr.py:347 Diagnostics still report required_author_associations as only ["OWNER"], even though allowlisted admin COLLABORATOR comments can now authorize. This is potentially misleading output, not a gate correctness issue.
• [P3] scripts/settle_tier4_pr.py:1521 CLI help still says --trusted-operator-login restricts MEMBER comments only; it should mention allowlisted COLLABORATOR admin settlement too.

dogfood: yes

[scarmani]

Tier-4 Human Settlement Authorization

PR: #8412
Exact head: 4850d5d
Authorized action: admin_squash_merge and branch_protection_reconcile, only if #8412 is non-draft and live exact-head checks/merge-packet remain otherwise green.

Human-risk settlement: I accept the Tier 4 risk for this PR.