diff --git a/docs/reviews/2026-11-17_RDM-developer-guide_PR103_Review_Records_r1.md b/docs/reviews/2026-11-17_RDM-developer-guide_PR103_Review_Records_r1.md
new file mode 100644
index 0000000..43bb7de
--- /dev/null
+++ b/docs/reviews/2026-11-17_RDM-developer-guide_PR103_Review_Records_r1.md
@@ -0,0 +1,37 @@
+# PR103 Review records r1 / レビュー記録 r1
+
+> **Summary:** Pass; the single added review is byte-identical to its original, and all 13 head checks succeeded.
+> **概要:** pass。追加されたレビュー1件は原本とバイト一致し、head の検査13件は全件成功。
+
+**Jump:** [English](#english) · [日本語](#japanese)
+
+Version / 版: 0.1 — 2026-11-17. **Learned:** The published copy preserves the original verdict and evidence limitations exactly. / 公開用の写しは原本の判定と検証範囲の限界を正確に保持している。
+
+Applies to: RDM-developer-guide PR #103, REV(担当レビュワ, Codex Astra)レビュー r1.
+Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Reviewed record](2026-11-17_RDM-osf.io-dev2027_PR47_Wiki_Abort_Queued_And_Last_Level_r1.md). Used by: [PR #103](https://github.com/RCOSDP/RDM-developer-guide/pull/103).
+
+---
+
+
+## English
+
+[→ 日本語版へ](#japanese)
+
+**Verdict: pass.** No findings. Reviewed head `83e6b1f2ccb899c018bdf534c0d2fd9f360ce0a3` against base `a686349eb7d614e7f489b74481362235a861caf4` in an independent clone. The only change adds the linked PR47 record (97 lines); `cmp` confirmed byte identity with the supplied original. SHA-256: `19ed9fdbba60e6e716b02f29da382cc72e6718bbfd924d5cce5f4c2df4fa9a57`.
+
+The PR body accurately describes the record, reviewed PR47 head and resolved P3 findings. Its future review-record addition and deployment checks are stated as future work or merge conditions. No private-repository URL, local absolute path, apparent secret or leftover draft wording was found in the addition or PR body. The record preserves the distinction between independently checked and coordinator-supplied evidence.
+
+Local documentation checks passed (292 files, 8 exempt, 71 warnings); manifest and private-URL checks passed with no findings. All 13 GitHub checks on the exact head completed successfully; the PR head remained unchanged. Deployment render/up/network-health checks were not rerun in this documentation review and remain the coordinator's merge gate.
+
+---
+
+
+## 日本語
+
+[→ To English](#english)
+
+**判定: pass。** 指摘なし。独立した clone で head `83e6b1f2ccb899c018bdf534c0d2fd9f360ce0a3` と base `a686349eb7d614e7f489b74481362235a861caf4` を比較しました。変更はリンク先の PR47 記録の追加1件(97行)だけで、指定原本とのバイト一致を `cmp` で確認しました。SHA-256: `19ed9fdbba60e6e716b02f29da382cc72e6718bbfd924d5cce5f4c2df4fa9a57`。
+
+PR 本文の記録説明、PR47 の対象 head、解決済み P3 指摘の説明は正確です。今回のレビュー記録の追加と実機検証は、今後の作業またはマージ条件として明示されています。追加文書と PR 本文に、非公開リポジトリの URL、ローカル絶対パス、明らかな秘密情報、下書きの残存表現は見つかりませんでした。記録は独立確認と coordinator 提供の根拠を区別しています。
+
+ローカル文書検査は成功(292ファイル、除外8件、警告71件)、manifest・非公開 URL 検査も指摘なしで成功しました。対象 head の GitHub 検査13件は全件成功し、PR head の不変も確認しました。render・up・ネットワーク health の実機検証は今回の文書レビューでは再実行しておらず、coordinator のマージ条件として残ります。
diff --git a/docs/reviews/2026-11-17_RDM-osf.io-dev2027_PR47_Wiki_Abort_Queued_And_Last_Level_r1.md b/docs/reviews/2026-11-17_RDM-osf.io-dev2027_PR47_Wiki_Abort_Queued_And_Last_Level_r1.md
new file mode 100644
index 0000000..16e7ef4
--- /dev/null
+++ b/docs/reviews/2026-11-17_RDM-osf.io-dev2027_PR47_Wiki_Abort_Queued_And_Last_Level_r1.md
@@ -0,0 +1,97 @@
+# PR47 Wiki cancellation review r1 / Wiki 取消レビュー r1
+
+> **Summary:** Final verdict: pass. Both P3 documentation findings are resolved; all nine CI checks succeeded.
+> **概要:** 最終判定は pass。説明の P3 指摘2件は解決済み、CI は全9件成功。
+
+**Jump:** [English](#english) · [日本語](#japanese)
+
+Version / 版: 0.1 — 2026-11-17. **Learned:** Cancellation rows survive startup, but exception handling differs between checkpoints. / 取消行は起動を越えて残るが、確認位置によって例外の扱いが違う。
+
+Applies to: RDM-osf.io-dev2027 PR #47, REV(担当レビュワ, Codex Astra)レビュー r1.
+Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md). Used by: issue #42 follow-up (repository reference only).
+
+---
+
+
+## English
+
+[→ 日本語版へ](#japanese)
+
+### Final verdict: pass
+
+No open P1/P2. The unchanged head and updated body were rechecked; both P3 findings are resolved. The coordinator supplied all nine CI checks SUCCESS and mergeable=MERGEABLE; no CI polling was performed. Reviewed head `1239e7ba48226a67005f525b2a4b11d344eb299a` against base `d95e0abfae4ac6a6846d23b6b1daf4c1d68c93b0`, the eight tests and PR body. No tests were rerun because setting up the legacy services exceeded a cheap focused check. The author's 459 passing Wiki tests, clean lint and six detected mutants remain author-reported evidence.
+
+### Findings
+
+**P3 R1-01 — Resolved in the PR body: eager explanation and exception limitation.** The first abort check at `views.py:1255` precedes the try block. With nonempty input, the old eager path raises out of `_get_md_content_from_wb` before reaching page-import exception handlers. Swallowed RuntimeError describes tests that mock earlier helpers, not the normal complete import path. The new guard itself is correct. The “確認の失敗” limitation likewise describes only checks within page-import handlers: early checks and the new final check at line 1042 propagate failures. Failure at the final check prevents search-update enqueueing and end-time recording, even after pages were written. The updated body clarifies these boundaries and the narrower commit-message claim; no final-check failure-injection test is present.
+
+**P3 R1-02 — Resolved in the PR body: queued-cancel guarantee qualified.** The fixed scenario is cancellation whose Stopped update completes before startup. Cleanup still saves ABORTED before updating the row (lines 1408–1423); startup between those writes may overwrite ABORTED, allowing fetches until Stopped becomes visible. If import finishes first, cleanup's status-filtered update cannot change Completed. Dispatch also precedes row binding (lines 959–962), so concurrent cleanup before binding can miss the task. These are existing non-atomic lifecycle windows, not grounds to enlarge this narrowly authorized fix. The updated body now lists both alongside the final-check/Completed window. The tracer test serializes cleanup before startup and does not exclude these races.
+
+### Focused assessment
+
+Binding uses get_or_create with defaults, preserving Stopped. The unique task_id and exact filter isolate other tasks. check_running_task can overwrite Stopped with Error only when more than one other Running row exists; it then raises before import work, so it does not silently restart the cancelled import. Nothing restores Running. Completed can overwrite Stopped at the documented final-check boundary.
+
+The pinned Celery fork's build_tracer defaults to non-eager tracing and writes STARTED before the function and SUCCESS afterward. The pinned django-celery-results 1.0.1 manager updates existing result status, so the startup overwrite also applies to django-db. The cache-backend spy validates this ordering, not broker/process/production-backend integration. The context manager restores store_result; addCleanup restores eager mode even on test failure. No leak from either change was found. Non-eager checks still read Celery unless the row already proves cancellation; the guard does not mask backend errors.
+
+The final branch retains partial child results, enqueues one search update, records the task-specific end time and returns. Earlier abort branches also return, preventing duplicate enqueueing. Cancellation after the last page is consistent with the chosen semantics and disclosed; it does not undo pages. Tests cover partial final-level cancellation, final-page cancellation, normal completion and task-ID isolation. The partial-level test asserts the exact index payload; the final-page test asserts only one enqueue. The updated body replaces “1–2 checks per page” with the actual fetching, replacement, creation, depth-entry and final checkpoints.
+
+Killed workers, the soft-time-limit caveat and a real-worker test remain issue #42 scope. The documentation corrections and all-green CI satisfy the pass gate; no code-blocking finding was identified.
+
+### Final validation supplied by the coordinator
+
+These results were supplied for the unchanged head, not independently rerun.
+
+```toon
+ci_success[9]: Analyze python, Analyze javascript-typescript, build-image, tests addons, tests website, tests api1_and_js, tests api2, tests api3_and_osf, tests private_security
+mergeable: MERGEABLE
+deploy_test_merge: bf14a685ee
+render_up_health_network_exit: 0
+acceptance_run: acc-20261117T015337Z-63ef29
+acceptance_passed: 18
+acceptance_failed: 0
+```
+
+Deployment acceptance does not establish real-worker Wiki cancellation coverage.
+
+---
+
+
+## 日本語
+
+[→ To English](#english)
+
+### 最終判定: pass
+
+未解決の P1/P2 はありません。head 不変と修正後の本文を再確認し、P3 指摘2件は解決済みと判断しました。coordinator 提供の最終結果は CI 全9件 SUCCESS、mergeable=MERGEABLE です。CI のポーリングはしていません。head `1239e7ba48226a67005f525b2a4b11d344eb299a` と base `d95e0abfae4ac6a6846d23b6b1daf4c1d68c93b0` の差分、追加8試験、PR本文を確認しました。旧環境のサービス構築は安価な限定確認に収まらないため、試験は再実行していません。Wiki 459件成功、lint 指摘なし、6ミュータント検出は著者報告の根拠です。
+
+### 指摘
+
+**P3 R1-01 — 本文で解決済み: eager の説明と例外の限界。** `views.py:1255` の最初の取消確認は try より前です。入力が空でない通常の旧 eager 経路は `_get_md_content_from_wb` から例外が外へ出て、ページ作成側の例外処理には到達しません。RuntimeError が握りつぶされる説明は、前段ヘルパーを差し替えた試験には当てはまりますが、通常の全経路には当てはまりません。新しい guard 自体は妥当です。「確認の失敗」もページ作成側の例外処理内だけの説明で、前段や新しい最終確認(1042)の失敗は外へ伝播します。最終確認の失敗では、ページを書き込んだ後でも検索更新の依頼と終了時刻の記録に到達しません。修正後の本文には、この境界とコミットメッセージの説明範囲が明記されています。最終確認の障害注入試験はありません。
+
+**P3 R1-02 — 本文で解決済み: 待ち行列の取消保証の条件。** 修正されるのは起動前に Stopped の更新まで済んだ取消です。取消処理は依然 ABORTED の保存後に行を更新します(1408–1423)。この間に起動すると ABORTED が上書きされ、Stopped が見えるまでページ取得が進む可能性があります。先に取り込みが完了すると、状態条件付きの取消更新は Completed を変更できません。また配信が行作成より先なので(959–962)、作成前の並行取消はタスクを見落とせます。既存の非原子的な処理の隙間であり、今回の限定修正を広げる理由にはしません。修正後の本文は最終確認と Completed の隙間に加え、この2つの隙間も明記しています。tracer 試験は取消完了後に起動する順序で、これらの競合がないことまでは証明しません。
+
+### 対象項目の評価
+
+行の結び付けは defaults 付き get_or_create で、既存の Stopped を保持します。一意の task_id と一致条件により別タスクの取消は混入しません。check_running_task は他に Running が2行以上あれば Stopped を Error にできますが、その場で例外を出して取り込み前に終了し、取消済みの取り込みを黙って再開しません。Running に戻す処理はありません。Completed が Stopped を上書きできる最終確認後の境界は記載済みです。
+
+固定版 Celery fork の build_tracer は既定で非 eager の trace を行い、関数実行前に STARTED、後に SUCCESS を保存します。固定版 django-celery-results 1.0.1 の manager は既存結果の状態を更新するので、django-db でも起動時の上書きは成立します。cache backend の spy はこの順序を確認しますが、broker・別プロセス・本番 backend の結合試験ではありません。with が store_result を戻し、addCleanup が失敗時も eager を戻します。この2変更による漏れは見つかりませんでした。非 eager では行が取消を証明しない限り Celery を引き続き読み、guard は backend 例外を隠しません。
+
+最終分岐は子階層の途中結果を保持し、検索更新を1回依頼して対象タスクの終了時刻を記録し、return します。前の取消分岐も return するので二重依頼になりません。最後のページ後の取消扱いは採用した意味と整合し、本文にも明記されています。ページを元に戻す意味ではありません。試験は最終階層途中、最終ページ、通常完了、task_id の分離を確認しています。階層途中では検索対象の厳密な一覧、最終ページでは依頼回数を確認します。修正後の本文は「ページごとに1〜2回」を削除し、取得・置換・作成・階層入口・最終確認の各位置を正しく列挙しています。
+
+強制終了 worker、soft time limit の留保、実 worker 試験は issue #42 に残ります。説明訂正と CI 全件成功により pass の条件を満たしています。コードを阻害する指摘はありません。
+
+### coordinator 提供の最終検証結果
+
+次は不変の head に対して共有された結果で、レビュワによる再実行ではありません。
+
+```toon
+ci_success[9]: Analyze python, Analyze javascript-typescript, build-image, tests addons, tests website, tests api1_and_js, tests api2, tests api3_and_osf, tests private_security
+mergeable: MERGEABLE
+deploy_test_merge: bf14a685ee
+render_up_health_network_exit: 0
+acceptance_run: acc-20261117T015337Z-63ef29
+acceptance_passed: 18
+acceptance_failed: 0
+```
+
+デプロイ受入成功は、実 worker による Wiki 取消の検証を意味しません。