diff --git a/.github/workflows/codeql-scan-dispatch.yml b/.github/workflows/codeql-scan-dispatch.yml
index afb51676be..b698f6154e 100644
--- a/.github/workflows/codeql-scan-dispatch.yml
+++ b/.github/workflows/codeql-scan-dispatch.yml
@@ -589,7 +589,7 @@ jobs:
             | select(.event == "pull_request")
             | select(.path == ".github/workflows/codeql-pr.yml")
             | select(.head_sha == $head)
-            | select(.status == "completed")
+            | select(.status == "completed" and .conclusion == "failure")
             | select([
                 .pull_requests[]?
                 | select(.number == $pr_number and .head.sha == $head and .base.ref == $base_ref and .base.sha == $base)
@@ -601,22 +601,22 @@ jobs:
             exit 1
           fi
 
-          while IFS= read -r required_job; do
-            required_language="$(jq -r '.language' <<<"$required_job")"
-            required_job_id="$(jq -r '.job_id | tostring' <<<"$required_job")"
-            expected_name="CodeQL compatibility analysis (${required_language})"
-            job="$(gh api "repos/${TARGET_REPOSITORY}/actions/jobs/${required_job_id}")"
-            job_identity="$(printf '%s' "$job" | jq -r --arg head "$HEAD_SHA" --arg name "$expected_name" --argjson run_id "$REQUIRED_RUN_ID" --argjson job_id "$required_job_id" '
-              select(.id == $job_id and .run_id == $run_id and .head_sha == $head)
-              | select(.name == $name)
-              | select(.status == "completed" and .conclusion == "failure")
-              | .id // empty
-            ')"
-            if [ "$job_identity" != "$required_job_id" ]; then
-              echo "::error::CodeQL wake rejected missing or ambiguous required job identity."
-              exit 1
-            fi
-          done < <(printf '%s' "$REQUIRED_JOBS" | jq -c '.[]')
+          run_jobs="$(gh api --paginate "repos/${TARGET_REPOSITORY}/actions/runs/${REQUIRED_RUN_ID}/jobs" | jq -s '{jobs: map(.jobs[]?)}')"
+          expected_failed_ids="$(printf '%s' "$REQUIRED_JOBS" | jq -c '
+            [ .[]
+              | select((.language | type) == "string")
+              | select(.language | test("^[a-z0-9-]+$"))
+              | select((.job_id | tostring) | test("^[1-9][0-9]*$"))
+              | .job_id | tonumber
+            ] | sort
+          ')"
+          observed_failed_ids="$(printf '%s' "$run_jobs" | jq -c '
+            [ .jobs[] | select(.status == "completed" and .conclusion == "failure") | .id ] | sort
+          ')"
+          if [ "$expected_failed_ids" != "$observed_failed_ids" ]; then
+            echo "::error::CodeQL wake rejected a run whose bound failed CodeQL jobs do not exactly match its failed jobs."
+            exit 1
+          fi
 
           gh api -X POST "repos/${TARGET_REPOSITORY}/actions/runs/${REQUIRED_RUN_ID}/rerun-failed-jobs" >/dev/null
           echo "Re-ran failed jobs in exact CodeQL run ${REQUIRED_RUN_ID} for ${HEAD_SHA} on base ${BASE_REF}@${BASE_SHA} after all dispatch shards completed."
diff --git a/AGENTS.md b/AGENTS.md
index ff01a0aecf..eef7ed2e34 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -60,6 +60,10 @@ The materialization contract is also covered by [`docs/doctoring/exact-artifact-
   head. If a current-head dispatch is cancelled while deduplicating, enqueue
   exactly one replacement for that PR and workflow and verify the replacement
   carries the same live target head.
+- A `repository_dispatch` handler executes its default-branch workflow, not an
+  open control-plane PR's file. Do not add `workflow_dispatch` merely to test a
+  privileged branch implementation. Use fixture-backed contracts before merge,
+  then verify the first default-branch dispatch after protected integration.
 - CodeQL language shards do not wake required jobs independently. After the
   complete `scan` matrix terminates, one `wake-required-codeql` coordinator
   revalidates the live PR and exact required run before one run-level
@@ -202,6 +206,11 @@ them alone proves succession.
   for every literal you touched — event-type strings, cron expressions, environment-variable
   names, tuple members, pinned digests — not only the obviously named sibling test. A change
   can satisfy one oracle and still leave a second, independent one stale.
+- For a multi-language CodeQL dispatch, do not rerun the required jobs from
+  matrix shards. After every shard has produced its terminal gate outcome,
+  verify that the original run's complete failed-job set exactly matches the
+  authenticated CodeQL binding, then issue one run-level failed-jobs rerun.
+  GitHub rejects the second concurrent job rerun with HTTP 403.
 - Read a stale pull request's own changes with a three-dot diff —
   `git diff <base>...<head>` — or with `gh pr diff`, which is already three-dot. A two-dot
   `git diff <base> <head>` renders everything the base gained since the fork point as though
@@ -225,7 +234,9 @@ them alone proves succession.
 - A successful rerun of one matrix job does not rerun its sibling matrix jobs. Therefore an
   `already running` response from a second per-job rerun must remain a failure: even if the shared
   run is active, that sibling can still retain its old failed verdict. Coordinate the wake only
-  after all dispatch shards publish, then rerun the exact run's failed jobs as one operation.
+  after all dispatch shards terminate, including a scan failure whose exact verdict must be
+  reflected by the required check; exclude cancellation because it has no complete evidence.
+  Then rerun the exact run's failed jobs as one operation.
 - A `codeql-dispatch/<language>` commit status is head-scoped and carries neither the PR base
   nor the required-run identity. Keep it as diagnostic output only. Shards and the coordinator
   may accept a terminal verdict only from a completed central dispatch run named with
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f2775e20ce..6f3ce3c2d5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -13,6 +13,9 @@
   while a sibling language scan is still running. The coordinator now keeps an
   active run with the same immutable repository, PR, head, base, and required
   run identity, avoiding same-PR cancellation of valid evidence.
+- The run-level wake starts after a terminal scan matrix whether its verdict is
+  success or failure, so exact failure evidence reaches the required check; a
+  cancelled matrix remains excluded because it has no complete verdict.
 
 ### Failed-check finding names the Strix sandbox instead of the gateway
 
@@ -69,6 +72,13 @@
 
 ### CodeQL scan dispatch matrix serialisation
 
+- Documented the CodeQL wake bootstrap boundary: a `repository_dispatch` run
+  executes the default-branch handler, so an open control-plane PR cannot
+  exercise its proposed privileged workflow by selecting its branch. The
+  pre-merge evidence is fixture-backed contracts and actionlint; the first
+  protected default-branch dispatch is the required live proof. No
+  `workflow_dispatch` bypass was added.
+
 - Serialised the dispatched CodeQL matrix with `toJSON()` in `codeql-scan-dispatch.yml`. `codeql-pr.yml` sends `client_payload.matrix` as an array and the handler assigned it straight into `env:`, where a value must be a scalar, so GitHub rejected the step with "A sequence was not expected" and the dispatched scan never ran -- 0 successes against 136 failures since the handler was added in #1776. The validate step already consumes the value through `jq`, so JSON text is the shape it was written for and no consumer changes. Added a string contract test, because neither `yaml.safe_load` nor `actionlint` 1.7.12 flags this: it is an Actions template rule, so only GitHub's own validator rejects it and no local gate catches the class.
 
 ### Contextual-orchestrator pin refresh
@@ -84,6 +94,11 @@
 - Raised `hourly-review-repair.yml`'s discovery ceiling from 50 to 200 while rotating deterministic 50-PR deep-inspection windows by hourly run number. The scheduler hydrates only the selected window and stops immediately after its single dispatch, preserving access to newer PRs without quadrupling expensive review/check/comment work. See `docs/doctoring/hourly-review-repair-single-file-consolidation.md`'s 2026-09-03 follow-up.
 
 ## [Unreleased]
+- Serialize multi-language CodeQL dispatch wakeups after the scan matrix has
+  completed. The owner now proves that the original exact-head run's entire
+  failed-job set equals the authenticated CodeQL job binding before issuing one
+  failed-jobs rerun, preventing the second shard from receiving GitHub's
+  `workflow run already running` 403.
 - Include merge-scheduler entrypoint, core, and regression-test changes in
   the existing runtime-quality workflow's trigger and suite selector. Scheduler
   workflow edits retain queue checks and also select the full review-repair
diff --git a/CLAUDE.md b/CLAUDE.md
index 115456763f..2e39d3fd13 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -18,6 +18,9 @@ configuring any such loop.
 The repo/Project — not private agent memory — is the source of truth. This file complements those
 documents; it does not replace them.
 
+`repository_dispatch` executes the default-branch workflow. An open central workflow PR therefore
+needs fixture-backed contracts before merge and a fresh default-branch dispatch after protected
+integration; never add branch-selected `workflow_dispatch` to bypass that boundary.
 For CodeQL's dispatch-and-wake loop, wait for the complete scan matrix, then let
 one `wake-required-codeql` coordinator revalidate the live PR and exact required
 run before one run-level failed-job rerun. The wake identity includes PR number,
@@ -144,6 +147,10 @@ repeatable compile command.
   needs the same evidence.
 - **100% coverage and 100% docstrings on `scripts/ci/`** are hard gates, not aspirations. New helper
   code needs matching tests and docstrings.
+- **Multi-language CodeQL dispatch wakeups are run-level, not shard-level.** Wait for every
+  matrix shard, prove the failed-job set equals the authenticated binding, then rerun failed jobs
+  once. Concurrent per-job reruns make the first request reactivate the run and GitHub rejects the
+  second with HTTP 403.
 - **Product hourly callers** stay thin. Do not hard-code OriginWeave, aFIPC, naruon, or Keyverse
   into `pr-review-fix-scheduler.yml`. The model credential remains `NVIDIA_NIM_API_KEY`
   on the worker, never `COPILOT_GITHUB_TOKEN`.
@@ -229,7 +236,8 @@ repeatable compile command.
   branch or from the autofix flow's conflict-marker resolution.
 - **Per-job reruns do not cover matrix siblings.** Do not accept a second shard's `already running`
   response merely because the shared run is active. Coordinate after every dispatch shard has
-  published its verdict and wake the exact run's failed jobs once, so no sibling retains a stale
+  terminated and wake the exact run's failed jobs once, so a scan failure is reflected rather than
+  suppressed and no sibling retains a stale
   failed required check. The wake is valid only while PR number, head SHA, and base SHA still match
   both live PR metadata and the exact required run's `pull_requests[]` association.
 - **Head-only CodeQL statuses are diagnostic, not terminal authority.** They cannot distinguish
diff --git a/docs/adr/0025-codeql-required-workflow-dispatch-architecture.md b/docs/adr/0025-codeql-required-workflow-dispatch-architecture.md
index 4a84c800f5..518e666212 100644
--- a/docs/adr/0025-codeql-required-workflow-dispatch-architecture.md
+++ b/docs/adr/0025-codeql-required-workflow-dispatch-architecture.md
@@ -191,8 +191,9 @@ forbidden.
 The 2026-09-05 rejection of "full matrix in one dispatch" is therefore
 superseded. Siblings are jobs in one run, not runs in one concurrency group.
 `required_jobs` remains a 1:1 map of language to canonical job id; the
-post-matrix wake validates the whole map and every exact job before one
-run-level rerun. A missing, stale, or mismatched identity fails closed. The
+post-matrix wake requires the whole map to equal the run's complete failed-job
+id set before one run-level rerun. A missing, stale, or mismatched identity
+fails closed. The
 old scalar `required_job_id`/`required_language` payload is retained only as
 a bounded queued-payload compatibility path where the matrix has exactly one
 language; it is not the current producer contract.
diff --git a/docs/doctoring/codeql-dispatch-multi-language-wake-race.md b/docs/doctoring/codeql-dispatch-multi-language-wake-race.md
new file mode 100644
index 0000000000..822dbc47fa
--- /dev/null
+++ b/docs/doctoring/codeql-dispatch-multi-language-wake-race.md
@@ -0,0 +1,51 @@
+# CodeQL dispatch multi-language wake race
+
+## Incident evidence
+
+On 2026-09-09, `.github` PR #1857 at
+`afeffe3b6a7a5494be1dae12322a0fc2a78c6efe` dispatched run `34320386978`
+after both initial CodeQL required jobs reported pending. Its `actions` scan
+completed successfully, but the subsequent Python wake failed with GitHub's
+`The workflow run containing this job is already running` HTTP 403. The first
+per-language rerun had already reactivated the shared required-workflow run.
+
+The scan itself was not the failing evidence: both the CodeQL analysis and the
+Medium-or-higher SARIF gate completed. A status publication 403 is separately
+recorded as non-terminal for a clean scan, as required by the existing status
+publication contract.
+
+## Corrective action
+
+`codeql-scan-dispatch.yml` now waits for the complete matrix and uses one
+run-level `rerun-failed-jobs` call. Before that mutation, it revalidates the
+open PR and exact head, validates the required workflow run identity, fetches
+all jobs, and refuses unless the full failed-job identifier set exactly equals
+the authenticated CodeQL binding. This preserves the no-unrelated-job
+invariant without relying on concurrent per-job reruns.
+
+## Verification and recovery
+
+`tests/test_codeql_scan_dispatch_workflow_contract.py` executes the wake
+block against fixture-backed GitHub responses. It covers the bounded job set,
+stale and closed pull requests, a mismatched or nonfailed job set, and an
+already-running required workflow. Hosted exact-head evidence remains required
+before any PR or consumer result is treated as successful. If the binding does
+not match, do not retry a job manually; inspect the exact run and dispatch a
+new current-head scan only through the owner workflow.
+
+### Bootstrap boundary
+
+On 2026-09-09, PR `#2056` at
+`aad55ed864db3466d52d24dcd88a80d15e84996d` triggered dispatch run
+`34322210652` for required run `34321725703`. The run executed the protected
+default-branch handler, as GitHub defines for `repository_dispatch`; its log
+therefore used the old per-job wake and reproduced the same HTTP 403 when the
+second shard tried to rerun an already-running required workflow. This does
+not execute or disprove #2056's proposed run-level wake.
+
+The branch cannot use `workflow_dispatch` as a substitute: that would let a
+caller select an unprotected workflow ref while minting privileged cross-repo
+credentials, and the central queue contract forbids it. The pre-merge proof is
+the fixture-backed contract plus actionlint. After protected integration, a
+fresh default-branch dispatch must prove the one-call wake against an exact
+current head before a consumer adopts the owner.
diff --git a/docs/doctoring/codeql-partial-shard-wake-duplicate-dispatch.md b/docs/doctoring/codeql-partial-shard-wake-duplicate-dispatch.md
index f5e5267e84..6bce1e3b9d 100644
--- a/docs/doctoring/codeql-partial-shard-wake-duplicate-dispatch.md
+++ b/docs/doctoring/codeql-partial-shard-wake-duplicate-dispatch.md
@@ -41,7 +41,7 @@ wake block을 fixture-backed `gh api`로 실행하면 두 경우 모두 return c
 - live PR은 open이고 `base.sha == BASE_SHA`, `head.sha == HEAD_SHA`여야 한다.
 - exact `REQUIRED_RUN_ID`는 pull_request event의 `codeql-pr.yml` completed run이며
   `pull_requests[]` 안에 같은 PR number/head/base ref/base SHA tuple이 정확히 하나 있어야 한다.
-- 그 뒤에만 기존 failed-job id/name/run/head 검증과 run-level
+- 그 뒤에만 complete bound failed-job-id 집합 검증과 run-level
   `rerun-failed-jobs`가 실행된다.
 
 같은 fixture를 repaired block에 적용하면 두 changed-base 경로 모두 return
diff --git a/docs/product-technical-gap-baseline.md b/docs/product-technical-gap-baseline.md
index 27a5539de9..688b8c312c 100644
--- a/docs/product-technical-gap-baseline.md
+++ b/docs/product-technical-gap-baseline.md
@@ -10,9 +10,10 @@
 같은 exact head의 CodeQL dispatch `34316388553`에서는 Python shard가
 성공한 뒤 required job을 깨웠고, Actions shard가 분석 중일 때 동일 제목의
 dispatch `34317266381`가 생성됐다. 같은 PR concurrency가 첫 실행을 취소해
-Actions SARIF가 사라졌다. 중앙 coordinator는 이제 repository·PR·head·base·
-required run id가 모두 같은 queued/running dispatch를 찾으면 재전송하지
-않는다. focused RED→GREEN 증거와 실행 시각은
+Actions SARIF가 사라졌다. wake는 모든 shard가 성공한 뒤 exact failed run을
+한 번만 깨우며, 중앙 coordinator는 repository·PR·head·base·required run
+id가 모두 같은 queued/running dispatch를 찾으면 재전송하지 않는다.
+focused RED→GREEN 증거와 실행 시각은
 `docs/doctoring/codeql-partial-shard-wake-duplicate-dispatch.md`에 남긴다.
 
 ## 1. 근거와 범위
diff --git a/tests/test_codeql_scan_dispatch_workflow_contract.py b/tests/test_codeql_scan_dispatch_workflow_contract.py
index 0c8c88f462..e90421b535 100644
--- a/tests/test_codeql_scan_dispatch_workflow_contract.py
+++ b/tests/test_codeql_scan_dispatch_workflow_contract.py
@@ -561,17 +561,17 @@ def test_dispatch_wakes_failed_jobs_once_after_all_language_shards() -> None:
     assert "BASE_SHA: ${{ needs.validate-dispatch.outputs.base_sha }}" in wake_job
     assert 'gh api "repos/${TARGET_REPOSITORY}/pulls/${PR_NUMBER}"' in wake
     assert 'gh api "repos/${TARGET_REPOSITORY}/actions/runs/${REQUIRED_RUN_ID}"' in wake
-    assert 'gh api "repos/${TARGET_REPOSITORY}/actions/jobs/${required_job_id}"' in wake
+    assert 'actions/runs/${REQUIRED_RUN_ID}/jobs' in wake
     assert 'select(.event == "pull_request")' in wake
     assert 'select(.path == ".github/workflows/codeql-pr.yml")' in wake
     assert "select(.head_sha == $head)" in wake
     assert ".base.ref == $base_ref" in wake
     assert ".base.sha == $base" in wake
-    assert ".run_id == $run_id" in wake
-    assert "select(.name == $name)" in wake
     assert 'select(.status == "completed" and .conclusion == "failure")' in wake
+    assert "bound failed CodeQL jobs" in wake
     assert 'actions/runs/${REQUIRED_RUN_ID}/rerun-failed-jobs' in wake
     assert 'actions/jobs/${required_job_id}/rerun"' not in wake
+    assert "while " not in wake
     assert "sleep " not in wake
 
 
@@ -656,13 +656,13 @@ def _run_wake_step(
         '  fi\n'
         "  exit 0\n"
         "fi\n"
+        'if [ "${2:-}" = "--paginate" ]; then\n'
+        '  printf \'%s\\n\' "$FAKE_JOBS_PAGE"\n'
+        "  exit 0\n"
+        "fi\n"
         'case "$2" in\n'
         '  */pulls/*) printf \'%s\\n\' "$FAKE_PULL_JSON" ;;\n'
         '  */actions/runs/*) printf \'%s\\n\' "$FAKE_RUN_JSON" ;;\n'
-        '  */actions/jobs/*)\n'
-        '    job_id="${2##*/}"\n'
-        '    jq -c --argjson job_id "$job_id" \'map(select(.id == $job_id)) | first // empty\' <<<"$FAKE_JOBS_JSON"\n'
-        '    ;;\n'
         "  *) exit 1 ;;\n"
         "esac\n",
         encoding="utf-8",
@@ -673,7 +673,7 @@ def _run_wake_step(
         "PATH": f"{fake_bin}:{os.environ['PATH']}",
         "FAKE_PULL_JSON": json.dumps(pull),
         "FAKE_RUN_JSON": json.dumps(run),
-        "FAKE_JOBS_JSON": json.dumps(jobs),
+        "FAKE_JOBS_PAGE": json.dumps({"jobs": jobs}),
         "FAKE_POST_LOG": str(post_log),
         "FAKE_RERUN_ERROR": rerun_error or "",
         "GH_TOKEN": "fake-token",
@@ -732,7 +732,7 @@ def test_dispatch_wake_rejects_stale_head_and_closed_pr(tmp_path: Path) -> None:
     assert not closed_log.exists()
 
 
-def test_dispatch_wake_rejects_ambiguous_or_nonfailed_job_identity(tmp_path: Path) -> None:
+def test_dispatch_wake_rejects_unbound_or_nonfailed_job_set(tmp_path: Path) -> None:
     wrong_job_result, wrong_job_log = _run_wake_step(
         tmp_path / "wrong-job",
         jobs=[
@@ -762,12 +762,13 @@ def test_dispatch_wake_rejects_ambiguous_or_nonfailed_job_identity(tmp_path: Pat
 
     assert wrong_job_result.returncode == 1
     assert successful_job_result.returncode == 1
-    assert "missing or ambiguous required job identity" in wrong_job_result.stdout
+    assert "bound failed CodeQL jobs" in wrong_job_result.stdout
     assert not wrong_job_log.exists()
     assert not successful_job_log.exists()
 
 
 def test_dispatch_wake_rejects_nonterminal_required_run(tmp_path: Path) -> None:
+    """The single wake runs only after the matrix has completed."""
     result, post_log = _run_wake_step(
         tmp_path,
         run={