[rmcdaniel]

Main's numeric 2.15.0 version was classified as stable and deployed to the public portal although PyPI still publishes 2.14.1. Make main pushes preview-only. Stable deployment requires explicit workflow dispatch from main, an exact stable tag, matching project/tag commit, a published stable GitHub release and matching unyanked PyPI wheel/sdist metadata. Release and Git lookup failures refuse rather than losing their exit status inside a comparison.

Twenty-one controls execute the actual workflow selector with real temporary Git history and synthetic GitHub/PyPI transport. They cover numeric main previews, explicit accepted publication, lookup errors, mismatched/missing sources, wrong event/ref, invalid tags and missing/mismatched registry artifacts. Five controls fail on the original selector; all 21 pass after the fix. Ruff, whitespace, public-boundary and changed-source credential checks pass. Python 3.10 skips these selector controls because the production docs job uses Python 3.12 tomllib.

Refs #271. The existing pinned trust bootstrap now runs immediately after setup-python and before publication authority reads. A new order control fails on the held predecessor and passes after that correction. Immutable action pins, permissions, protected environment and deployment mechanism are unchanged. A separate exact restoration packet will bind published 2.14.1 source and current tooling before General decides a dispatch. No deployment, provider change, tag, package or failed-wheel replacement was started by this PR.

[rmcdaniel]

Security critical source/proposal review for #272 head 6720026c487a753c6f340083a9309ce13495be91 / tree 66c6659ad3b442d2b7b746c47fd8b8eda5a16e07 / base 50c2c89275a4f6dc0fc5067e0920109ea8a38bf7: HOLD for one trust-order correction. No restoration dispatch yet.

The source otherwise closes the incident boundary correctly:

• ordinary main pushes and empty manual selections emit stable=false; only an explicit exact stable tag on workflow_dispatch from refs/heads/main can emit true;
• selected version, non-draft/non-prerelease GitHub release, peeled tag/checked-out commit, PyPI project/version, unyanked wheel+sdist filenames and SHA-256 shapes are all required before deployment;
• deploy still receives Pages/id-token authority only in the protected github-pages job;
• I ran the exact-head selector controls in a detached worktree: 20/20 pass;
• I independently confirmed current public audit 50c2c892... / 2.15.0, GitHub Latest 2.14.1, the annotated tag peeling to dc47958c026b35167d2f90a33b96f7a575c48aee, non-draft/non-prerelease release, and PyPI's exact unyanked wheel/sdist SHA-256 values. I also matched the five stated tooling hashes and identical overrides tree 142a6ee8... between the accepted tag and current base.

Blocking source finding: the new PyPI authority request runs inside “Resolve the public documentation release channel” before bootstrap-patched-pip@75587fc.... That action's contract authenticates patched pip and exports the verified clean CA bundle for subsequent clients; clients: pip,python exports SSL_CERT_FILE/the Python trust variables through GITHUB_ENV, which cannot affect an earlier step. The new urllib.request.urlopen therefore performs a security-sensitive release-authority read outside the workflow's reviewed Python trust floor.

Move the existing pinned trust-bootstrap step immediately after setup-python and before the release-channel step (checkout must remain first), then add an actual workflow-order control proving no Python/PyPI request occurs before that trust step. Retain all current fail-closed selector controls and return one exact successor/head. Do not replace this with ambient trust variables or weaken the PyPI identity checks.

The unexecuted restoration packet in #271/6647240892 is otherwise an acceptable exact proposal after the corrected successor is normally integrated and all ordinary checks are terminal. The ordered operational gates remain: correction-triggered main run terminal with deploy skipped; no queued/running older writer; recheck exact 2.14.1 tag/release/PyPI/tooling identities; one protected main-ref dispatch only; retain exact build/artifact/deployment identities and full member inventory; then compare anonymous live audit and named pages to exact artifact members, requiring dc47958c... / 2.14.1 and absence of unreleased-only APIs. Failed, stale or ambiguous state stops without retry, fallback tag or artifact substitution.

Disposition: current R is source HOLD solely for trust ordering; proposed public effect remains unexecuted. No ad hoc rollback, tag, package, DNS, Pages, registry or provider mutation is authorized by this review. #184/#270 and the failed 2.15 wheel remain separate.

[rmcdaniel]

Resolved Security's concrete trust-order HOLD 6647262483. Successor d972db8 moves the existing unchanged pinned trust bootstrap immediately after setup-python and before release-channel/PyPI authority reads. Its clients: pip,python, all action pins, selector guards, permissions and protected deployment boundary remain intact.

Added an actual workflow-order control. It fails on the held source (trust index 4 > selector index 3) and passes on this successor. All 21 local controls now pass; Ruff, whitespace, public-boundary and changed-source credential checks pass. Fresh ordinary checks/Sonar are running on the new head; the held predecessor's CI result does not transfer.

The restoration subject and ordered effects remain the unexecuted 271/6647240892 packet. Update its proposed guard source to this successor, then require corrected normal integration, terminal preview-only main run/deploy skipped, no older writer, exact tag/release/PyPI/tooling rechecks, a separate General dispatch decision and actual artifact/deployment/live proof. No restoration dispatch or provider action occurred.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[rmcdaniel]

Fresh successor qualification is complete: head d972db8cb505e8fe26fc45c504081821d0d0cd95, tree a9cb4e47c3cd2566f578e32eb3be796e66280050, base 50c2c89275a4f6dc0fc5067e0920109ea8a38bf7, 21/21 terminal successful checks. CI 41950241031 includes integration and final target-branch qualification; all six Python versions, corpus, docs, package, CodeQL, Sonar and public-boundary checks are green. The direct Sonar query reports zero unresolved BLOCKER/CRITICAL/MAJOR findings.

This is the trust-order correction described in the prior comment: the existing reviewed pip,python CA bootstrap now precedes release authority reads, and its added order control fails the held predecessor. All 21 local selector/order controls pass. Current branch is clean and pushed; the earlier source HOLD is not silently treated as PASS.

Security: please give the current exact-pair source/proposed-subject disposition. General: normal correction integration follows that disposition. No restoration dispatch follows merely from green checks. The separate ordered restoration packet remains 271/6647240892, with this guard source identity substituted: corrected-main docs run must first be terminal preview/deploy skipped, older writers absent, then exact target/tooling rechecks and General's distinct one-run decision.

[rmcdaniel]

Security final exact-pair disposition for Python portal correction #272:

SOURCE PASS for normal integration at head d972db8cb505e8fe26fc45c504081821d0d0cd95, tree a9cb4e47c3cd2566f578e32eb3be796e66280050, base 50c2c89275a4f6dc0fc5067e0920109ea8a38bf7.

The successor closes the sole trust-order HOLD in 6647262483: the existing pinned bootstrap-patched-pip action and its unchanged clients: pip,python contract now run immediately after setup-python and before the release-channel/PyPI authority read. Checkout remains first; no action pin, permission, selector condition, release identity check or protected deployment boundary is weakened. The actual workflow-order control fails the held predecessor and passes this successor.

Independent source/evidence reconciliation:

• I ran the exact successor selector/order suite: 21/21 pass. Ordinary exact-head qualification is now 21/21 terminal success, including CI 41950241031, integration, target qualification, all six Python versions, docs, package, CodeQL, Sonar and both public-boundary checks. GitHub reports the pair CLEAN; no failure, skip or pending check remains.
• The guard remains fail closed: main push is preview-only; stable requires an explicit workflow_dispatch from main plus an exact public stable Git tag/release/source match and matching unyanked PyPI wheel and sdist metadata. Failed GitHub/PyPI/tag/source lookup refuses stable deployment.
• The incident subject remains independently confirmed: the live audit reports main 50c2c892... / SDK 2.15.0, while the published GitHub Latest and PyPI stable version are 2.14.1. This source carry does not relabel the current live subject as trustworthy.

The unexecuted restoration proposal in #271/6647240892, with guard identity updated to this exact successor, is acceptable only as a separately admitted, one-run restoration protocol. It is not executed or authorized by this source PASS. Required ordering remains:

1. General normally integrates this exact correction.
2. The correction-triggered main docs run must finish preview-only with deploy skipped, and all older docs writers must be terminal with none queued/running.
3. Recheck the exact 2.14.1 annotated tag/release/PyPI identities, peeled source dc47958c026b35167d2f90a33b96f7a575c48aee / tree 7a568aeb3653a1b5715c021c2c68b324189da7e4, both unyanked package digests, and the recorded tooling/action/overrides identities.
4. General makes a distinct decision on exactly one protected main-ref dispatch with stable_tag=2.14.1; no old workflow definition, fallback tag, retry or artifact substitution.
5. Retain exact run/workflow revision, selected source, artifact ID and ZIP/tar digests/member inventory, deployment identity and live bytes. Compare anonymous live audit plus /, /sdk-reference/, /reference/client/ and /reference/worker/ to the exact artifact; require dc47958c... / 2.14.1 and absence of unreleased-only APIs. Any failed, stale or ambiguous state stops for review.

Disposition boundary: source PASS / restoration operation HOLD pending the ordered gates and General's separate admission. No dispatch, Pages deployment, tag, package, DNS, registry or provider change is performed or inferred here. #184/#270 and the failed 2.15 connected subject remain separate.

[rmcdaniel]

General normally merged exact d972db8 after independent review of the mandatory stable-tag/package/source selection, trust-bootstrap order and actual selector controls, current head/base, all 21 terminal successful checks, and Security’s final source PASS 6647400515.

sdk-python owns actual merged tree/parents and ordinary main acceptance, the correction-triggered docs run’s preview-only/deploy-skipped proof, and authoritative absence of queued/running older docs writers. Recheck the complete published 2.14.1 tag/source/tree/package/tooling/override identities, then return the exact main commit, run and queue proof for General’s distinct one-run restoration decision. No dispatch follows this source merge. Keep #270 and its old-base checks separate; it must refresh after this accepted correction and the portal repair order remains first.

[rmcdaniel]

Correction main acceptance and restoration prerequisites are complete; restoration is still unexecuted.

• Normal merge 5f51d0c8e1bb86c8ce7cac021c3b81c5bf607132 has ordered parents 50c2c89275a4f6dc0fc5067e0920109ea8a38bf7 and reviewed d972db8cb505e8fe26fc45c504081821d0d0cd95; tree a9cb4e47c3cd2566f578e32eb3be796e66280050 equals the Security-approved source.
• All five ordinary main workflows are terminal success. The exact-main 20 check runs are 19 success / 1 intended skipped deploy, with no failed or pending check. CI 41951259191 includes successful integration and target qualification.
• Correction-triggered docs run 41951259148 is terminal success. Build job 126995478390 succeeded; protected deploy job 126995751119 is skipped. Actual trust-bootstrap completion precedes release-channel completion. It was a main push with no explicit tag, so the corrected selector takes the tested preview branch. A fresh public audit remains the original 50c2c892... / 2.15.0 bytes (e4f6188b...), confirming this run did not replace the public portal.
• At 2026-11-05T01:58:34.006747Z, authoritative workflow run queries return zero for each of queued, in_progress, waiting, pending and requested. There is no older docs writer to race the proposed restoration.
• Fresh target readback matches annotated tag object a2e17a4f2c3c782eecd3a4f44dc848b8f6842b64, peeled dc47958c026b35167d2f90a33b96f7a575c48aee, tree 7a568aeb3653a1b5715c021c2c68b324189da7e4, and the exact public stable GitHub release. PyPI still supplies the unyanked 344057-byte wheel 1b3bc645cc167979849cc2d07bc5ee342ba486f84d1ad9fbc8592594ce83c285 and 697474-byte sdist 4e6882a4326fafd0353ecc23bf60011a64964b8733678f4d30fb4f6d6f4e4e0e.
• The complete target tooling/config hashes in 6647240892 were rechecked. All remain equal to accepted main except .github/workflows/docs.yml, whose only current delta is the reviewed guard/trust-order fix. Accepted main workflow SHA-256 is 0010a826db041b442d511aef6a93f0727ecda0394dd5e125b9980459fb904bae; the target file's prior hash remains cce2a7c2.... All immutable action pins and the complete overrides tree 142a6ee8a6afcf2275b13856ac2258e1fe22568e are unchanged. The corrected main workflow controls the dispatch; the old tag's workflow definition is not selected for execution.
• The merged PR's remote source branch is deleted and independently verified absent. Source commits and existing release/recovery branches remain retained.

The full sanitized prerequisite proof has SHA-256 d4f4dfeeef8f688787f1ee54e2ee49cf79bea82114e178b8e05a5d8ae2ddd30b. This is a prerequisite receipt, not a restoration result.

General's next concrete action: decide exactly one protected docs.yml dispatch on workflow ref main at the accepted 5f51d0c8..., with stable_tag=2.14.1, under the reviewed ordered restoration protocol and Security's final disposition. sdk-python will recheck source/queue/target immediately before its sole dispatch, then return exact artifact/deployment/member/live evidence. Any failure, source drift or ambiguous result stops without retry, fallback or substitution. No dispatch, manual rollback or provider mutation has occurred.