docs(gap): scope Actions fan-out evidence and capacity conclusions

[seonghobae]

Summary

Stacked on #1903 because both PRs update docs/product-technical-gap-baseline.md; this remains the existing canonical ledger lane rather than a duplicate PR.

The 35-workflow static analysis and its trigger/matrix/required-context observations are preserved, but the conclusion is scoped to what the method can establish:

• 33 is a conservative static admission ceiling/estimate for the modeled synchronize event, not observed concurrent runner occupancy and not a measured lower bound.
• The analysis did not identify additional waste in the modeled trigger set; it is not an exhaustive proof that recoverable workflow waste is absent.
• fix(scheduler): reject commented reviews before runner admission #1900 exact head 17eb37df91c4c5ce5beaeb2f243f79af45d17150 removed a verified 56-second runner-held wait and 105 lines from one workflow path. Its branch evidence reports 470 CI tests, 3 behavior tests, and official actionlint passing, but fix(scheduler): reject commented reviews before runner admission #1900 is not yet protected-main evidence.
• Required-context-preserving structural waste removal and org-level capacity investigation are parallel levers. Neither evidence substitutes for the other.

This PR only updates the product/technical gap ledger. It does not weaken required checks, claim an organization-wide queue reduction, or treat approval, permissions, local tests, or an open successor as protected integration.

Current stacked authority

• exact head: fcdb8dfe6951704cce688ec2e3756837b04bd71c
• current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903 parent: f4ff7f8c025c4d0a15145c3cd634d96c92326ec3
• reconciliation: two-parent, non-force; docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903 is an ancestor, behind_by=0
• scope against current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903: one file, 149 additions
• Ready is review admission. Hosted terminal GREEN and qualifying independent review remain merge gates, not Ready prerequisites.

Test plan

• Exact-head tests/test_product_technical_gap_baseline.py: 5 passed
• Three-dot git diff --check: clean
• Exact tree/parents and one-file stacked scope verified
• Residual literal job\\namong formatting defect replaced with a real Markdown line break
• Qualifying independent review
• Protected integration through ordinary branch protection

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: db4d7bbb-695c-461a-b14d-228d9f603375

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

[P2] 확인된 낭비와 모순되는 최종 감사 결론을 수정해 주세요

검토 HEAD는 f7f480a820b1a8278eb4537943b0fcf830b9e97c입니다. docs/product-technical-gap-baseline.md의 이 PR 추가 절은 두 번의 정정 뒤에도 첫머리의 “No recoverable waste was found”와 마지막 “Optimising .github's workflow YAML further is not a productive lever”, “measured floor ... 33”을 유지합니다. 이 문구는 다음 작업자가 아직 필요한 구조 감사를 중단하게 만듭니다.

직접 확인한 반례는 다음과 같습니다.

• 본문이 인용한 #1910은 실제로 2026-09-05T08:44:26Z에 c4a89b31a02caa1f1a5be1c36d54252cb8e57f8b로 병합됐습니다. 보호 context를 없애지 않고 echo-only 직렬 의존성을 줄일 수 있었으므로, YAML 변경의 실효성이 없다는 결론과 양립하지 않습니다. 이 조회는 병합 사실 확인이며 후속 실행 성능이나 정상 보호 병합 절차까지 입증하지 않습니다.
• #1899는 서로 다른 main SHA의 Strix 실행 33928897846, 33932271770, 33933530334가 08:28:26–08:45:01 UTC에 겹친 실제 job 근거와, 첫 branch-push를 run-ID로 분리하던 식의 수정 전/후 검증을 갖고 있습니다. PR feature-branch의 push 중복이 없다는 결과는 이런 main 연속 push 누적이나 구형 run 재실행의 최신 검증 취소를 반증하지 않습니다. #1899는 아직 미병합이므로 적체 해소 완료로 계산하지 않습니다.
• unknown 조건을 실행으로 세는 정적 합계는 해당 분석 범위의 보수적 상한 추정입니다. 실제 runner 동시 점유 수나 줄일 수 없는 하한이 아닙니다. 한 synchronize의 정적 job 개수만으로 재실행·지연 도착·dispatch 경쟁·직렬 큐 대기까지 소진했다고 말할 수 없습니다.

최소 수정은 기존 관측표를 유지하면서 제목·첫 문단·마지막 결론을 “관측한 trigger 범위와 남은 구조 감사”로 좁히는 것입니다. runner 용량 확인과 workflow 낭비 제거는 병행할 수 있습니다. 정적 분석기와 실행 명령이 PR에 없으므로 33이라는 수치는 재현 가능한 산출물을 연결하기 전에는 독립 검증된 하한으로 쓰지 말아 주세요. 현재 #1899의 tests/test_review_rerun_concurrency.py 등 실제 YAML 식 회귀와 #1910의 의존성 변경을 함께 대조해야 합니다.

이는 사용자 요청을 더 작은 목표로 바꾸지 않기 위한 문서 정확성 finding입니다. org 설정·보안 gate 변경이나 추가 모델 timeout을 요청하지 않습니다.

2026-09-07 추가 근거 — 같은 HEAD 재확인

현재 #1905는 여전히 f7f480a820b1a8278eb4537943b0fcf830b9e97c이며 위 결론이 본문 2815·2941행에 남아 있습니다. 보호 main c9052e607e5f3cc76e73207e7786b21500721b79의 scheduler publication 대기는 최대 8회 조회 사이 2+4+6+8+10+12+14=56초 동안 runner를 잡고 있었습니다. 해당 review 이벤트 경로는 auto-merge를 하지 않으며, 기존 opencode-review-dispatch.yml의 승인 게시 후 scheduler 단계가 실제 후속 병합 검증을 소유합니다. 기존 PR #1900 head 17eb37df91c4c5ce5beaeb2f243f79af45d17150에서 이 대기 105줄을 제거했고 최신 base 통합 후 470개 CI 검사·3개 행동 검사와 공식 actionlint가 통과했습니다. 검증 기록.

이는 해당 경로의 불필요한 대기 제거 근거이며 실제 조직 적체 감소·모든 경로 56초 단축·보호 병합을 증명하지는 않습니다. #1900의 현재 hosted 검사는 아직 대기 중입니다. 기존 분석 수치와 capacity 미확인 상태는 보존하되 추가 구조 수리가 무의미하다는 일반화는 수정해야 합니다.

[opencode-agent]

Pull request overview

OpenCode could not approve from deterministic current-head evidence because GitHub Checks have failed.

Findings

1. HIGH Current-head GitHub Checks - Fix failed required checks before approval

• Problem: Failed same-head checks remain for f7f480a820b1a8278eb4537943b0fcf830b9e97c.
• Root cause: The model-unavailable evidence fallback is allowed only when peer GitHub Checks are complete and clean.
• Fix: Read and fix the failed check logs below, then rerun the current-head checks.
• Regression test: Keep the model-unavailable fallback gated on an empty failed-check rollup.

Failed checks:

• Required Noema Review/noema-review: FAILURE (https://github.com/ContextualWisdomLab/.github/actions/runs/33955810883/job/101306378575)
• noema-review check run: failure (https://github.com/ContextualWisdomLab/.github/actions/runs/33955810883/job/101306378575)

Changed-File Evidence Map

flowchart LR
  PR["PR changed files"] --> Evidence["OpenCode bounded evidence"]
  Evidence --> S1["Docs: product-technical-gap-baseline.md"]
  S1 --> I1["operator or user guidance"]
  I1 --> R1["Review risk: Docs: product-technical-gap-baseline.md"]
  R1 --> V1["docs review"]

Loading

[opencode-agent]

OpenCode Review Overview

• Head SHA: fcdb8dfe6951704cce688ec2e3756837b04bd71c
• Workflow run: 34098851316
• Workflow attempt: 1
• Gate result: REQUEST_CHANGES
• Coverage gate: success
• Model pool: exhausted
• Verdict: REQUEST_CHANGES
• Formal review: docs(gap): scope Actions fan-out evidence and capacity conclusions #1905 (review)

[seonghobae]

Classification for the two failing checks on f7f480a8: the noema-review failure OpenCode cites (run 33955810883) was created on the morning of 2026-09-05, i.e. it executed the pre-#1939 review sidecar whose serving catalog was NVIDIA-only — today's structural gateway class, not a defect in this diff — and opencode-review cannot approve while a same-head check has failed (its own fallback is gated on an empty failed-check rollup) and its verdict dispatch is in any case blocked by the .github#1929 actor variable. A re-run keeps the creation-time workflow_sha; only a new push re-binds the current sidecar (#1939 → #1947 → #1949 have since changed the catalog and preflight). Note also that this PR's base is not main, so the pipeline will not merge it as-is regardless; retargeting to main (after merging main into the head) is the owner's call. Not pushing here.

[seonghobae]

추가 실측 후보를 공유합니다. YAML 최적화가 더는 유효하지 않다는 결론은 범위를 좁혀야 합니다. 기존 #1962는 Strix의 read-only changed-scope/admit 작업을 하나의 runner로 합쳐 metadata job 수를 2→1로 줄입니다(로컬 검증 완료, 아직 미병합이며 org queue 감소 실증은 아님). 현재 c31aa0234dcae6bf1b7e27c55be7bdc87d4fe507의 security-scan.yml:97–102, sast-semgrep.yml:74–79, strix.yml:179–184에는 3회 파일목록 조회가 모두 실패하면 마지막 시도 뒤에도 sleep 9초가 실행됩니다. 세 번째 뒤에는 추가 조회가 없으므로 그 대기는 재시도에 기여하지 않습니다. 앞선 3초/6초 backoff와 불완전 목록일 때 전체 스캔 정책은 보존해야 합니다. 이는 해당 실패 경로에서 작업당 9초 절감 후보이지, 조직 60-job ceiling의 주원인이나 정상 경로의 절감 증거가 아닙니다. required context를 삭제하거나 보안 gate를 약화하지 않고도 가능한 국소 개선과 조직 capacity 문제를 구분해 주세요.

[seonghobae]

기존 원장/분석을 보존한 채 결론 범위만 좁히는 owner coordination입니다. 새 baseline/원장 PR을 만들지 않습니다.

Fresh live state:

• PR #1905는 여전히 f7f480a820b1a8278eb4537943b0fcf830b9e97c, open/mergeable, target docs/gap-baseline-item4-clean입니다.
• 기존 finding #5550986403이 지적한 본문 결론(No recoverable waste was found, Optimising .github's workflow YAML further is not a productive lever, measured floor ... 33)은 current head 기준으로 아직 유효한 수정 대상입니다.
• #1900은 여전히 open/unmerged이며 exact head 17eb37df91c4c5ce5beaeb2f243f79af45d17150입니다. 이 lane은 scheduler의 runner-held publication wait 105줄을 제거했고, handoff finding에 기록된 최신 검증은 470 CI contracts + 3 behavior tests + official actionlint GREEN입니다. 이것은 protected integration이나 org-wide queue reduction 증거로 승계하면 안 됩니다.

수정 원칙:

1. 기존 35-workflow static analysis, ruleset/required-context 관측, queue/capacity 관측표와 33이라는 정적 계산 자체는 삭제하지 않습니다.
2. 단, unknown을 실행으로 세는 33은 해당 event/model의 보수적 static admission ceiling/estimate이지, 실제 동시 runner occupancy나 줄일 수 없는 measured floor가 아닙니다.
3. #1900의 실제 repair는 적어도 한 review/scheduler 경로에서 불필요한 runner-held wait가 제거 가능함을 보여 주므로 No recoverable waste / not a productive lever 같은 전역 부정 결론은 더 이상 유지할 수 없습니다.
4. 따라서 결론을 static ceiling ≠ observed concurrent occupancy ≠ measured lower bound로 명시하고, capacity investigation과 구조적 workflow waste 제거가 병행 가능한 것으로 좁혀 주세요.
5. #1900의 56-second observation은 해당 publication-poll path의 관측치일 뿐 모든 run이 56초 절감된다는 뜻이 아니며, 470+3/actionlint도 source verification이지 protected merge/hosted acceptance가 아닙니다.

같은 docs/product-technical-gap-baseline.md를 건드리는 PR이 다수이므로 current writer branch에서만 non-force로 수리하고 새 복제 PR/원장을 만들지 마십시오. 기존 finding #5550986403의 분석과 새 evidence를 함께 유지해 주세요.

[seonghobae]

Exact-head conclusion repair — 1f48d1635cffd24542c461395b5eb3f9a23265a3.

A concurrent writer first advanced the branch to 841e87fc… with the substantive conclusion narrowing; that intent was preserved. The final follow-up removes the remaining number never under-reports overclaim: treating unknown predicates as admitted prevents omission only within the statically enumerable model, not across dynamic/runtime behavior.

Current source no longer contains No recoverable waste was found, not a productive lever, measured floor is essentially, or number never under-reports. The 35-workflow observations and historical corrections remain intact. Exact-head tests/test_product_technical_gap_baseline.py: 5 passed; three-dot git diff --check: clean.

The PR remains a Ready stacked documentation PR on #1903. Prior-head CHANGES_REQUESTED and checks are not approval for this head; qualifying independent review and protected integration remain pending.

[seonghobae]

Exact-head handoff after non-force push guard stopped a duplicate local write.

• Local unpublished candidate db4f435a was not pushed because the remote branch moved; no force update was attempted.
• Remote 841e87fc7c7147589b8f8d91ca4b5854a60c37fd already carries the substantive static→bounded correction: 33 is described as a conservative static admission ceiling/estimate for the modeled trigger set, explicitly not observed concurrent runner occupancy and not a measured floor; the earlier global “no workflow waste” conclusion is narrowed, with fix(scheduler): reject commented reviews before runner admission #1900 preserved as counter-evidence.
• Remote successor/current head 1f48d1635cffd24542c461395b5eb3f9a23265a3 further removes the residual absolute claim “never under-reports”, replacing it with the bounded statement that the static model does not omit a potentially admitted job among paths it can enumerate.
• Therefore the unpublished source wording delta is already fully carried by the remote writer. Do not replay/cherry-pick db4f435a merely to duplicate this correction.

Reusable verification receipt from the unpublished local tree, provided as non-authoritative supporting evidence only: full repository suite 2994 passed, 1 skipped, 21 subtests passed, 297.46s, exit 0; documentation regression guard also passed. Receipt identifier: 1962f162. Because the remote head advanced after that local run, this is not current-head hosted evidence and must not be promoted as such; the existing writer may reuse the test shape or rerun it at the exact current head.

Current API head observed during this handoff: 1f48d1635cffd24542c461395b5eb3f9a23265a3. Devin Review and CodeRabbit commit statuses are currently success, while the PR’s stated exact-head hosted/documentation checks and ordinary protected integration remain separate requirements.

[seonghobae]

Exact-head stacked reconciliation — fcdb8dfe6951704cce688ec2e3756837b04bd71c.

• Root cause: the PR targeted branch name docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903, but its prior merge base was e40f71b28799b0d2f22dbcc89e5a35676cb73e9a; it was 24 commits behind current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903 even though GitHub still rendered a one-file three-dot PR.
• Repair: created a two-parent commit with prior docs(gap): scope Actions fan-out evidence and capacity conclusions #1905 head 1f48d1635cffd24542c461395b5eb3f9a23265a3 first and current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903 head f4ff7f8c025c4d0a15145c3cd634d96c92326ec3 second. The branch advanced with force=false; concurrent writer intent remains in history.
• GitHub comparison from current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903: behind_by=0, one changed path, docs/product-technical-gap-baseline.md, 149 additions.
• Exact remote tree aa1b8a62d3f4ad24c14f81020c4052e4a871ccd1: documentation contract 5 passed; three-dot diff check clean; parent list and one-file scope verified.
• The residual literal job\\namong introduced on the predecessor head was detected during reconciliation and replaced with a real line break.
• Review state: zero inline threads. The formal CHANGES_REQUESTED review is bound to predecessor f7f480a8… and is not dismissed or treated as current approval. Its only recorded blocker was the old-head Noema failure; current hosted and qualifying independent review evidence remain pending.

New exact-head hosted runs are queued, not GREEN:

• Security Scan: https://github.com/ContextualWisdomLab/.github/actions/runs/34078849481
• CodeQL PR: https://github.com/ContextualWisdomLab/.github/actions/runs/34078849437
• SAST Semgrep: https://github.com/ContextualWisdomLab/.github/actions/runs/34078849468

No auto-merge, approval, bypass, empty push, manual rerun, force push, or review dismissal was performed.

[seonghobae]

Ready-for-review admission restored on unchanged exact head fcdb8dfe6951704cce688ec2e3756837b04bd71c.

Fresh final audit immediately before the transition found the PR had been converted back to Draft even though:

• current docs(gap-baseline): root-cause item 4's provider-pool stall, 8 data points #1903 f4ff7f8c025c4d0a15145c3cd634d96c92326ec3 is a parent and comparison is behind_by=0;
• the effective delta is one documentation path;
• exact documentation contracts are 5 passed and diff check is clean;
• inline review threads are 0;
• the only formal CHANGES_REQUESTED review is bound to predecessor f7f480a820b1a8278eb4537943b0fcf830b9e97c and reports that predecessor's Noema check failure, not a remaining current-source finding.

Ready event: 2026-09-07T03:16:43Z. Ready admits current-head review; it is not approval, merge, hosted GREEN, or protected integration.

The Ready event cancelled the earlier queued runs and registered new same-head runs, all still queued:

• CodeQL PR: https://github.com/ContextualWisdomLab/.github/actions/runs/34079111710
• SAST Semgrep: https://github.com/ContextualWisdomLab/.github/actions/runs/34079111737
• Security Scan: https://github.com/ContextualWisdomLab/.github/actions/runs/34079111714

No auto-merge, approval, bypass, empty push, manual rerun, force push, or review dismissal was performed.

[opencode-agent]

Pull request overview

OpenCode could not approve from deterministic current-head evidence because GitHub Checks have failed.

Findings

1. HIGH Current-head GitHub Checks - Fix failed required checks before approval

• Problem: Failed same-head checks remain for fcdb8dfe6951704cce688ec2e3756837b04bd71c.
• Root cause: The model-unavailable evidence fallback is allowed only when peer GitHub Checks are complete and clean.
• Fix: Read and fix the failed check logs below, then rerun the current-head checks.
• Regression test: Keep the model-unavailable fallback gated on an empty failed-check rollup.

Failed checks:

• Required Noema Review/noema-review: FAILURE (https://github.com/ContextualWisdomLab/.github/actions/runs/34079112045/job/101622342955)
• noema-review check run: failure (https://github.com/ContextualWisdomLab/.github/actions/runs/34079112045/job/101622342955)

Changed-File Evidence Map

flowchart LR
  PR["PR changed files"] --> Evidence["OpenCode bounded evidence"]
  Evidence --> S1["Docs: product-technical-gap-baseline.md"]
  S1 --> I1["operator or user guidance"]
  I1 --> R1["Review risk: Docs: product-technical-gap-baseline.md"]
  R1 --> V1["docs review"]

Loading

[seonghobae]

The failing check is not in this PR's diff — and a rerun is not currently justified

The review asks to "read and fix the failed check logs, then rerun the current-head checks." I read them. There is nothing in this PR to fix, and I am deliberately not rerunning yet. Evidence for both halves:

1. The diff cannot cause this failure

This PR changes one file, docs-only:

modified  +149  -0   docs/product-technical-gap-baseline.md

The failing job (101622342955) dies in the step Prepare Noema model verdict, with the real annotation (source-echo lines filtered out):

##[error]Noema gateway transport failed: HTTPError: HTTP Error 502: Bad Gateway;
         caller attempts=1, duration=1357.7s, phase=response_error,
         served_model=google/gemma-4-31b-it

A 502 from the gateway after a 22.6-minute wait, on attempts=1. A markdown edit cannot produce that.

2. Why I am not rerunning right now

noema-review completions across 14 repositories, last 6 hours:

most recent SUCCESS   2026-09-07T05:39:59Z   (5.7 h ago, inkspan)
most recent FAILURE   2026-09-07T07:33:20Z   (3.8 h ago, seedream_evasepic)
everything since      cancelled (no model call reached)

No successful Noema review in 5.7 hours. Both decisive observations since are failures, and a rerun I triggered on .github#1687 at 10:07Z failed the same way — 502, attempts=1, after a 117.9-minute queue wait. Rerunning here would spend another queue slot plus 13–24 minutes of runner time to very probably reproduce the same 502.

I will rerun as soon as there is evidence the gateway serves again — a single noema-review success anywhere is enough.

Context: this lane's failures are expensive

Measured across 66 repositories, 503 completed runs (jobs that actually held a runner):

concl      n    total min   median   max
failure   98      2492.2     22.3   101.3
success   90       923.1      9.8    27.1

Failures are 52% of completions but 69% of the lane's runner time — a failing job holds a runner ~2.28× as long as a passing one, because the 502 arrives only after the full wait. That is why blind reruns here are costly rather than free, and why route-health evidence is worth checking first.

Ratio and share are window-independent; the absolute minutes are not comparable across repos, since each contributed its newest page.

🤖 Generated with Claude Code

[seonghobae]

Canonical baseline handoff for user-requested review/agent skill adoption: #2012, exact head 166a649c5de96f14cc4f1b06d44ee66c4b23df0d, is stacked on #1655@5cbce30b4905e033480d4a9c7f8195be7f029bc9 and remains Draft/unmerged. This lane did not overwrite your baseline source or dated inventory.

Please append the following bounded entry through this existing baseline owner (not a duplicate baseline PR):

Gap ID	Observed gap / implemented delta	Evidence	Remaining acceptance
G-REVIEW-SKILL-PROJECTION-20260907	External skills were discoverable but did not reach isolated reviewers. #2012 adds one instruction-only native skill and identical projections into the two root prompts already copied by trusted OpenCode dispatch; selects Ponytail, Addy Agent Skills and Scientific Agent Skills without hooks/installers/permission or provider changes.	#2012; Proposed ADR docs/adr/20260907_review_skill_projection.md; source/commit/blob/license register bundled under .agents/skills/cwl-review-evidence/references/; local packaging RED 2 fail/1 pass → GREEN 3 pass, not LLM evaluation.	Parent integration; non-force restack; child exact-head checks and independent review; protected prompt-consumption receipt; source-bound model comparison; Noema/Strix/vendor/workstation owner integrations. No live rollout, release or org-wide installation claim.

The root baseline, AGENTS/CLAUDE and CHANGELOG source have not been updated by this child. Its new procedure is Unreleased. Preserve all other lane deltas, controls and evidence distinctions.

[opencode-agent]

Pull request overview

OpenCode could not approve from deterministic current-head evidence because GitHub Checks have failed.

Findings

1. HIGH Current-head GitHub Checks - Fix failed required checks before approval

• Problem: Failed same-head checks remain for fcdb8dfe6951704cce688ec2e3756837b04bd71c.
• Root cause: The model-unavailable evidence fallback is allowed only when peer GitHub Checks are complete and clean.
• Fix: Read and fix the failed check logs below, then rerun the current-head checks.
• Regression test: Keep the model-unavailable fallback gated on an empty failed-check rollup.

Failed checks:

• Required Noema Review/noema-review: FAILURE (https://github.com/ContextualWisdomLab/.github/actions/runs/34079112045/job/101622342955)
• noema-review check run: failure (https://github.com/ContextualWisdomLab/.github/actions/runs/34079112045/job/101622342955)

Changed-File Evidence Map

flowchart LR
  PR["PR changed files"] --> Evidence["OpenCode bounded evidence"]
  Evidence --> S1["Docs: product-technical-gap-baseline.md"]
  S1 --> I1["operator or user guidance"]
  I1 --> R1["Review risk: Docs: product-technical-gap-baseline.md"]
  R1 --> V1["docs review"]

Loading