ContextualWisdomLab · seonghobae · Sep 6, 2026 · Sep 7, 2026 · Sep 7, 2026 · Sep 7, 2026
@@ -109,8 +109,19 @@ concurrency:
     github.event.pull_request.number ||
     github.event.client_payload.pr_number ||
     (github.event_name == 'push' && format('push-{0}', github.ref_name)) ||
-    github.run_id }}
-  cancel-in-progress: true
+    github.run_id }}-${{
+    github.event.action == 'closed' && github.run_id ||
+    github.event_name == 'push' && 'protected-ref' ||
+    github.event.pull_request.head.sha ||
+    github.event.client_payload.pr_head_sha || github.run_id }}
+  # Draft/Ready and duplicate admission events can share one exact head. Do
+  # not let those lifecycle events destroy an executing scanner verdict. The
+  # metadata-only cleanup job below remains the cancellation owner for a
+  # verified superseded head or closed pull request. Closed events use their
+  # unique run id above so cleanup cannot queue behind the scan it must stop.
+  # Only a newer protected-branch push cancels in-progress work; PR lifecycle
+  # events use exact-head identities and never stop same-head provider work.
+  cancel-in-progress: ${{ github.event_name == 'push' }}
 
 # Scorecard Token-Permissions (alert #43): keep the workflow-level token
 # read-only and scope same-repo status publication to the Strix scan job.
@@ -242,8 +253,10 @@ jobs:
 
   cancel-superseded-pr-runs:
     if: >-
-      github.event_name == 'pull_request_target' &&
-      (github.event.action == 'synchronize' || github.event.action == 'converted_to_draft' || github.event.action == 'closed')
+      (github.event_name == 'pull_request_target' &&
+      (github.event.action == 'synchronize' || github.event.action == 'closed')) ||
+      (github.event_name == 'repository_dispatch' &&
+      github.event.client_payload.pr_number != '')
     # Idempotent per PR: a fresh sweep re-verifies live state (live_target_matches
     # below) before selecting or cancelling anything, so it fully subsumes
     # whatever an older, not-yet-run instance would have done. cancel-in-progress
@@ -257,7 +270,9 @@ jobs:
     concurrency:
       group: >-
         cancel-superseded-pr-runs-${{
+        github.event.client_payload.target_repository ||
         github.event.pull_request.base.repo.full_name || github.repository }}-${{
+        github.event.client_payload.pr_number ||
         github.event.pull_request.number || github.run_id }}
       cancel-in-progress: true
     runs-on: ubuntu-24.04
@@ -277,10 +292,11 @@ jobs:
       pull-requests: read
     env:
       GH_TOKEN: ${{ secrets.PR_REVIEW_MERGE_TOKEN || secrets.OPENCODE_APPROVE_TOKEN || github.token }}
-      TARGET_REPOSITORY: ${{ github.event.pull_request.base.repo.full_name || github.repository }}
-      TARGET_PR_NUMBER: ${{ github.event.pull_request.number }}
-      TARGET_PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
-      PR_ACTION: ${{ github.event.action }}
+      RUN_REPOSITORY: ${{ github.repository }}
+      TARGET_REPOSITORY: ${{ github.event.client_payload.target_repository || github.event.pull_request.base.repo.full_name || github.repository }}
+      TARGET_PR_NUMBER: ${{ github.event.client_payload.pr_number || github.event.pull_request.number }}
+      TARGET_PR_HEAD_SHA: ${{ github.event.client_payload.pr_head_sha || github.event.pull_request.head.sha }}
+      PR_ACTION: ${{ github.event_name == 'repository_dispatch' && 'synchronize' || github.event.action }}
       CURRENT_RUN_ID: ${{ github.run_id }}
     steps:
       - name: Cancel queued and running scans for superseded or inactive pull requests
@@ -289,18 +305,16 @@ jobs:
           set -euo pipefail
 
           live_target_matches() {
-            local live_pr_json live_state live_draft live_head
+            local live_pr_json live_state live_head
             if ! live_pr_json="$(gh api "repos/${TARGET_REPOSITORY}/pulls/${TARGET_PR_NUMBER}" 2>/tmp/strix-cleanup-gh-error)"; then
               echo "::warning::Strix cleanup could not verify the live pull request; leaving runs unchanged."
               sed 's/^/  /' /tmp/strix-cleanup-gh-error >&2 || true
               return 1
             fi
             live_state="$(jq -r '.state // ""' <<<"$live_pr_json")"
-            live_draft="$(jq -r '.draft // false' <<<"$live_pr_json")"
             live_head="$(jq -r '.head.sha // ""' <<<"$live_pr_json")"
             [ "$live_head" = "$TARGET_PR_HEAD_SHA" ] && {
               { [ "$PR_ACTION" = "closed" ] && [ "$live_state" = "closed" ]; } ||
-                { [ "$PR_ACTION" = "converted_to_draft" ] && [ "$live_state" = "open" ] && [ "$live_draft" = "true" ]; } ||
                 { [ "$PR_ACTION" = "synchronize" ] && [ "$live_state" = "open" ]; }
             }
           }
@@ -311,7 +325,7 @@ jobs:
               echo "::notice::Strix cleanup target changed before run selection; leaving runs unchanged."
               return 0
             fi
-            local runs_url="repos/${TARGET_REPOSITORY}/actions/runs?status=${status}&per_page=100"
+            local runs_url="repos/${RUN_REPOSITORY}/actions/runs?status=${status}&per_page=100"
             local runs_json
             if ! runs_json="$(gh api --paginate "$runs_url" 2>/tmp/strix-cleanup-gh-error)"; then
               echo "::warning::Strix cleanup could not inspect ${TARGET_REPOSITORY}; leaving runs unchanged."
@@ -320,25 +334,26 @@ jobs:
             fi
             local run_ids
             if ! run_ids="$(jq -r --arg pr "$TARGET_PR_NUMBER" --arg head_sha "$TARGET_PR_HEAD_SHA" \
-              --arg action "$PR_ACTION" --arg repo "$TARGET_REPOSITORY" --arg current "$CURRENT_RUN_ID" '
+              --arg action "$PR_ACTION" --arg repo "$TARGET_REPOSITORY" --arg run_repo "$RUN_REPOSITORY" \
+              --arg current "$CURRENT_RUN_ID" '
                 .workflow_runs[]
                 | select((.id | tostring) != $current)
                 | select(.name == "Strix Security Scan")
-                | select(.event == "pull_request_target")
+                | select(.event == "pull_request_target" or .event == "repository_dispatch")
+                | (($run_repo | ascii_downcase) == ($repo | ascii_downcase)) as $metadata_is_target_repository
                 | ((.display_title // "") | startswith("Strix Security Scan " + $repo + "#" + $pr + "@")) as $title_matches
-                | ((.pull_requests // []) | any((.number | tostring) == $pr)) as $metadata_matches
+                | ($metadata_is_target_repository and ((.pull_requests // []) | any((.number | tostring) == $pr))) as $metadata_matches
                 | select($title_matches or $metadata_matches)
                 | ((.display_title // "") | endswith("@" + $head_sha)) as $title_is_current
-                | ((.pull_requests // []) | any(
+                | ($metadata_is_target_repository and ((.pull_requests // []) | any(
                     ((.number | tostring) == $pr)
                     and ((.head.sha // "") | ascii_downcase) == ($head_sha | ascii_downcase)
-                  )) as $metadata_is_current
-                | ((.pull_requests // []) | any(
+                  ))) as $metadata_is_current
+                | ($metadata_is_target_repository and ((.pull_requests // []) | any(
                     ((.number | tostring) == $pr) and ((.head.sha // "") != "")
-                  )) as $metadata_has_head
+                  ))) as $metadata_has_head
                 | select(
                     $action == "closed"
-                    or $action == "converted_to_draft"
                     or (($title_matches or $metadata_has_head) and (($title_is_current or $metadata_is_current) | not))
                   )
                 | .id
@@ -352,11 +367,11 @@ jobs:
                 echo "::notice::Strix cleanup target changed before cancellation; leaving runs unchanged."
                 return 0
               fi
-              if gh api --method POST "repos/${TARGET_REPOSITORY}/actions/runs/${run_id}/cancel" >/dev/null 2>/tmp/strix-cleanup-cancel-error ||
-                gh api --method POST "repos/${TARGET_REPOSITORY}/actions/runs/${run_id}/force-cancel" >/dev/null 2>>/tmp/strix-cleanup-cancel-error; then
-                echo "Cancelled obsolete Strix run ${run_id} in ${TARGET_REPOSITORY} for PR #${TARGET_PR_NUMBER}."
+              if gh api --method POST "repos/${RUN_REPOSITORY}/actions/runs/${run_id}/cancel" >/dev/null 2>/tmp/strix-cleanup-cancel-error ||
+                gh api --method POST "repos/${RUN_REPOSITORY}/actions/runs/${run_id}/force-cancel" >/dev/null 2>>/tmp/strix-cleanup-cancel-error; then
+                echo "Cancelled obsolete Strix run ${run_id} in ${RUN_REPOSITORY} for ${TARGET_REPOSITORY} PR #${TARGET_PR_NUMBER}."
               else
-                echo "::warning::Strix cleanup could not cancel run ${run_id} in ${TARGET_REPOSITORY}; it may have finished or the credential lacks Actions write access."
+                echo "::warning::Strix cleanup could not cancel run ${run_id} in ${RUN_REPOSITORY}; it may have finished or the credential lacks Actions write access."
                 sed 's/^/  /' /tmp/strix-cleanup-cancel-error >&2 || true
               fi
             done <<<"$run_ids"
@@ -367,8 +382,13 @@ jobs:
           done
 
   strix:
-    needs: [changed-scope, admit-current-head]
-    if: needs.changed-scope.outputs.code == 'true' && needs.admit-current-head.outputs.admitted == 'true'
+    needs: [changed-scope, admit-current-head, cancel-superseded-pr-runs]
+    if: >-
+      always() && !cancelled() &&
+      needs.changed-scope.outputs.code == 'true' &&
+      needs.admit-current-head.outputs.admitted == 'true' &&
+      (needs.cancel-superseded-pr-runs.result == 'success' ||
+      needs.cancel-superseded-pr-runs.result == 'skipped')
     # Large, actively-growing repositories (e.g. contextual-orchestrator) can
     # legitimately require well over two hours to scan -- this org's own
     # standing operating directive accepts that central OpenCode/Strix/Noema

@@ -1,3 +1,26 @@
+### Strix reruns bind the exact PR base as well as the head
+
+- The scheduler now rejects a failed Strix job whose native pull-request
+  association belongs to an older base SHA, and revalidates both live base and
+  head immediately before the rerun mutation. Retargeting an unchanged head can
+  no longer replay an old-base scanner job as current evidence.
+
+### Strix preserves PR evidence and retires superseded push scans
+
+- Workflow-level `cancel-in-progress` is true only for `push`; Draft/Ready and
+  duplicate same-head PR admission events cannot destroy an executing scanner
+  verdict. The PR group is exact-head scoped, so a synchronized new head can start
+  its metadata-only superseded-run cleanup without waiting behind the old scan;
+  its provider job now waits for that cleanup to finish. A closed event uses its
+  unique run id for the same reason. Draft transitions preserve the current scan,
+  while the live-revalidated cleanup job covers both native and dispatched PR
+  runs and cancels only verified superseded heads or a closed pull request.
+  Native PR metadata is accepted only when the run and target repositories match,
+  preventing same-number cross-repository cancellation. No provider deadline or
+  merge-gate relaxation was added. This repairs the cancellation pattern
+  seen in runs `34068478185`, `34067942252`, and PR #1999 run `34067362987`,
+  while preserving #1938's protected-ref push coalescing and cancellation.
+
 ### Failed-check finding names the Strix sandbox instead of the gateway
 
 - `opencode-review-dispatch.yml`'s `emit_strix_provider_failure_finding` rendered one fixed finding for every `STRIX_PROVIDER_UNAVAILABLE` line, whose Root cause read "The contextual-orchestrator gateway or its discovered provider pool was unavailable for this run". `#1953` had just given the Strix sandbox bootstrap failure its own second verdict token (`STRIX_SANDBOX_UNAVAILABLE`) precisely because that attribution is wrong for it -- the sandbox container never reaches its Caido proxy, so the run dies before the gateway serves anything -- and this consumer re-applied the wrong attribution one step downstream, into the review findings and the failure census. The emitter now branches on the second token: a sandbox verdict gets a finding that names Strix's sandbox, says the verdict does not name the gateway, and tells the reader not to change gateway or provider configuration on its strength. A `STRIX_PROVIDER_UNAVAILABLE` line without the token keeps its existing text verbatim, so the gateway class has no regression surface. No test covered this finding text at all before (`gateway or its discovered provider pool` matched nothing under `tests/`); `tests/test_opencode_dispatch_strix_sandbox_finding.py` now runs the production emitter from the published run block and pins both directions plus the no-signal case. Refs #1953, #1935.

@@ -0,0 +1,48 @@
+# Scheduler의 명시적 OPEN·현재 head 확인
+
+## 원인과 범위
+
+#1902의 후속 조사에서 CodeQL 복구 primitive보다 먼저 고칠 공통 결함을 확인했다.
+`4bf80b99b6908c0323ac406d7e30e8346e09a50d`의
+`scripts/ci/pr_review_merge_scheduler_core.py`는 GraphQL 공통 PR fragment에서
+`state`를 요청하지 않았고, REST PR 정규화에서도 그 필드를 보존하지 않았다.
+그런데 `live_dispatch_head_matches`는 누락되거나 빈 state를 OPEN으로 취급했다.
+단일 PR 조회는 닫힌 PR도 반환하므로 head가 그대로면 닫힌 PR을 허용할 수 있었다.
+또한 양쪽 head를 빈 문자열로 대체해 비교했으므로 빈 값끼리도 일치했다.
+
+영향 범위는 OpenCode repository dispatch, Strix의 기존 job rerun,
+Strix repository dispatch 직전의 공통 guard다. 이번 수정은 이 세 경로의
+새 실행 요청을 막는 조건만 다룬다. 앞서 수행되는 stale-run cleanup의 순서나
+cancellation 정책은 바꾸지 않는다.
+
+## 수정
+
+- GraphQL 공통 fragment가 PR state를 실제로 요청한다.
+- REST fallback은 원본 state를 대문자로 보존하고, 누락은 빈 값으로 남긴다.
+- guard는 정확히 한 PR, 명시적 `OPEN`, 양쪽의 문자열 타입 40자리 hex SHA,
+  대소문자를 제외한 동일 head를 모두 요구한다.
+- 기존 정상 fixture는 `OPEN`을 명시한다. 누락 사례를 정상 fixture로 대체하지 않는다.
+
+토큰, 권한, trigger, queue, concurrency, dispatch payload는 변경하지 않았다.
+CodeQL primitive도 추가하지 않았다. 조회 직후 PR 상태가 바뀔 수 있는 경쟁 조건과
+중복 전송의 원자성은 여전히 미해결이며, 이 guard는 exact-once 보장이 아니다.
+Cross-repo target callback의 Actions-write 권한도 별도 미해결 조건이다.
+
+## 회귀 검증
+
+`tests/test_scheduler_live_dispatch_guard.py`는 실제 guard와 세 caller를 실행하고
+외부 API 및 실행 요청만 대체한다. 누락·빈 값·CLOSED·MERGED·UNKNOWN은 dispatch와
+rerun에 도달하지 않아야 하며, OPEN의 정상 경로는 계속 도달해야 한다.
+별도 사례가 빈 값, 잘못된 길이, 비-hex, 비문자열, 서로 다른 SHA를 거부하고
+GraphQL 실제 query와 REST fallback의 state 전달을 확인한다.
+
+Production 수정 전 새 회귀는 17 failed / 19 passed였다. 이후 실제 live head만
+잘못된 사례 두 건도 추가했다. 최종 관련 5파일은 `-W error`를 적용해 정상 환경에서
+380 passed, `GITHUB_ACTIONS=true` 환경에서도 380 passed를 확인했다.
+검증 명령은 다음과 같다.
+
+```sh
+python -m pytest -q -W error tests/test_scheduler_live_dispatch_guard.py tests/test_pr_review_merge_scheduler.py tests/test_strix_rerun_job_selection.py tests/test_repository_branch_coverage_review_schedulers.py tests/test_pr_review_fix_scheduler_rest_workflow_identity.py
+```
+
+로컬 회귀 통과는 실제 GitHub dispatch, protected merge, 대상 job 복구의 증거가 아니다.
@@ -72,6 +72,16 @@ burst each new head cancels the previous scan; the burst's final head is
 scanned, and the weekly full-tree `schedule` scan (unique run id, never
 cancelled) is the floor under a sustained burst.
 
+**Amendment (2026-09-07).** The workflow group now carries two independent
+identities. Pull-request work includes the exact head, while `closed` uses a
+run-unique suffix; with cancellation disabled for PR events, Ready, Draft, and
+same-head dispatch admission preserve an executing verdict, while a new head
+or closed cleanup does not wait behind it. Push work remains grouped by
+protected ref and is the only event class with `cancel-in-progress` authority.
+The replacement provider waits for live-revalidated cleanup to finish, and
+that cleanup enumerates both native and dispatched PR runs. No elapsed-time
+condition can cancel provider work.
+
 ## Verification
 
 - `python -m pytest -q tests/test_pr_review_merge_scheduler.py -k 'startup_failures or startup_failure'`