diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index f7654e96..c511f5ac 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -165,16 +165,6 @@ jobs:
     if: needs.detect.outputs.ready == 'true'
     uses: ./.github/workflows/run_tests.yml
 
-  # Deliberately absent from `release`'s `needs`. A failure here would skip tagging while
-  # the Release PR is already merged carrying `autorelease: pending`, so every later push
-  # would stand down and releases would stay wedged until someone cleared it by hand.
-  test-integration:
-    name: Test (integration)
-    needs: detect
-    if: needs.detect.outputs.ready == 'true'
-    uses: ./.github/workflows/run_integration.yml
-    secrets: inherit
-
   # Irreversible half.
   release:
     name: 🚀 Tag and release
diff --git a/.github/workflows/run_integration.yml b/.github/workflows/run_integration.yml
index a6ffcfa3..15154d4e 100644
--- a/.github/workflows/run_integration.yml
+++ b/.github/workflows/run_integration.yml
@@ -3,11 +3,11 @@ on:
   workflow_call:
     secrets: { }
 
-# A constant, not `github.workflow`, which inside a reusable workflow resolves to the
-# caller's name: the daily run and the pre-tag run would land in separate lanes and hit the
-# shared app at once. Queue rather than cancel, so the schedule cannot kill a release gate.
 concurrency:
-  group: run-integration-${{ github.ref }}
+  # Repository-wide, on purpose. The contended resource is the Stream app, not the branch,
+  # so keying on github.ref would let a *.x release run beside the daily run on the default
+  # branch. Not cancel-in-progress: a half-run leaves users and channels behind.
+  group: run-integration
   cancel-in-progress: false
 
 env:
diff --git a/.github/workflows/scheduled_test.yml b/.github/workflows/scheduled_test.yml
index 5f7ba0e3..3a589382 100644
--- a/.github/workflows/scheduled_test.yml
+++ b/.github/workflows/scheduled_test.yml
@@ -14,10 +14,15 @@ jobs:
     secrets: inherit
 
   # A failure here lands on no PR, and GitHub emails it only to whoever last edited the cron.
+  # A schedule event carries no repository in its payload, so it passes by event name; it
+  # always runs on the default branch anyway.
   report:
     name: Report a red run
     needs: integration
-    if: failure()
+    # Not failure(): a job that hits timeout-minutes concludes `cancelled`, and a hung run
+    # must report too. Default branch only: a manual dispatch on a feature branch must not
+    # touch the default branch's issue.
+    if: ${{ !cancelled() && needs.integration.result != 'success' && (github.event_name == 'schedule' || github.ref_name == github.event.repository.default_branch) }}
     runs-on: ubuntu-latest
     permissions:
       issues: write
@@ -28,8 +33,13 @@ jobs:
           TITLE: Daily integration run is red
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
         run: |
-          num="$(gh issue list --repo "$GITHUB_REPOSITORY" --state open --search "in:title \"$TITLE\"" \
-            --json number,title --jq "map(select(.title == \"$TITLE\")) | .[0].number // empty")"
+          # Unguarded, a non-2xx would abort the step under `bash -e` and a red run would
+          # report nothing at all. Fail toward creating the issue, which is the loud way.
+          if ! num="$(gh issue list --repo "$GITHUB_REPOSITORY" --state open --search "in:title \"$TITLE\"" \
+            --json number,title --jq "map(select(.title == \"$TITLE\")) | .[0].number // empty")"; then
+            echo "::warning::Could not list open issues; creating a new one."
+            num=""
+          fi
           if [ -n "$num" ]; then
             gh issue comment "$num" --repo "$GITHUB_REPOSITORY" --body "Still red: $RUN_URL"
           else
diff --git a/DEVELOPMENT.md b/DEVELOPMENT.md
index 8544dcd0..aa40b09c 100644
--- a/DEVELOPMENT.md
+++ b/DEVELOPMENT.md
@@ -44,7 +44,7 @@ make test-prometheus   # requires getstream[telemetry] deps
 | --- | --- | --- |
 | Pull request | `run_tests.yml`: ruff, ty, and `-m "not integration"` on five Python versions | yes, `🧪 Tests` is the required check |
 | Daily at 09:00 UTC | `run_integration.yml`: `-m integration`, both credential sets | no |
-| Push to `main` with a release pending | both; only the unit lane gates the tag | unit yes, integration no |
+| Push to `main` with a release pending | the unit lane | yes, it gates the tag |
 
 `@pytest.mark.integration` means one thing: the test talks to a live Stream app. The unit
 lane therefore runs with no credentials, no `environment:` and no `STREAM_*`. Keep it that
@@ -52,9 +52,9 @@ way. A fork PR gets no secrets and still goes green, and a live test added witho
 marker fails in CI instead of quietly passing on someone else's credentials.
 
 Integration gates nothing, anywhere. It runs against an app five SDK repos share, so
-another repo's run or a backend regression can redden it with nothing wrong here, and a red
-pre-tag run used to wedge every later release behind `autorelease: pending`. A red daily run
-opens an issue titled "Daily integration run is red". Fix it, do not route around it.
+another repo's run or a backend regression can redden it with nothing wrong here. It does not
+run before a tag either. A red daily run opens an issue titled "Daily integration run is
+red". Fix it, do not route around it.
 
 A Release PR skips the lane and `🧪 Tests` passes in seconds on a `skipped` result. The
 merge commit still runs it before the tag.