diff --git a/.github/workflows/monitoring.yml b/.github/workflows/monitoring.yml
index f3b485356..a411cf22f 100644
--- a/.github/workflows/monitoring.yml
+++ b/.github/workflows/monitoring.yml
@@ -48,7 +48,11 @@ jobs:
       - name: Measure publication heartbeat
         id: heartbeat
         run: |
+          # GitHub's default shell already sets -e, so a non-zero exit would
+          # abort before the code is recorded and the delivery step would be
+          # skipped — the check would detect the outage and tell no one.
           set -uo pipefail
+          set +e
           python3 scripts/publish_heartbeat.py \
             --repository-root . \
             --summary-output "$RUNNER_TEMP/heartbeat.md"
@@ -85,10 +89,15 @@ jobs:
 
       - name: Enforce divergence and stale thresholds
         id: release_state
+        if: always()
         env:
           MAIN_SHA: ${{ github.sha }}
         run: |
+          # GitHub's default shell already sets -e, so a non-zero exit would
+          # abort before the code is recorded and the delivery step would be
+          # skipped — the check would detect the outage and tell no one.
           set -uo pipefail
+          set +e
           main_sha="$(git rev-parse HEAD)"
           main_committed_at="$(git show -s --format=%cI HEAD)"
           python3 scripts/production_monitor.py \
@@ -139,10 +148,15 @@ jobs:
       # so it has to be caught here rather than hours later via staleness.
       - name: Check recent release runs
         id: release_runs
+        if: always()
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
+          # GitHub's default shell already sets -e, so a non-zero exit would
+          # abort before the code is recorded and the delivery step would be
+          # skipped — the check would detect the outage and tell no one.
           set -uo pipefail
+          set +e
           python3 scripts/check_release_runs.py \
             --repository "$GITHUB_REPOSITORY" \
             --summary-output "$RUNNER_TEMP/release_runs.md"
@@ -180,10 +194,15 @@ jobs:
       # scheduled work rather than an outage.
       - name: Measure capacity headroom
         id: capacity
+        if: always()
         env:
           PYTHONPATH: ${{ github.workspace }}
         run: |
+          # GitHub's default shell already sets -e, so a non-zero exit would
+          # abort before the code is recorded and the delivery step would be
+          # skipped — the check would detect the outage and tell no one.
           set -uo pipefail
+          set +e
           python3 scripts/capacity_report.py \
             --summary-output "$RUNNER_TEMP/capacity.md" \
             --fail-on never > "$RUNNER_TEMP/capacity.json"
diff --git a/tests/test_monitoring_alert_wiring.py b/tests/test_monitoring_alert_wiring.py
new file mode 100644
index 000000000..a5e854de8
--- /dev/null
+++ b/tests/test_monitoring_alert_wiring.py
@@ -0,0 +1,70 @@
+"""Wiring invariants for the monitoring job's alert delivery.
+
+Run 30798868809 detected a real outage — the publication heartbeat correctly
+exited 1 after 101 hours without a new post — and told nobody. GitHub's default
+shell already sets -e, so the non-zero exit aborted the step before
+`exit_code` reached $GITHUB_OUTPUT, and the delivery step's
+`steps.<id>.outputs.exit_code != ''` guard was therefore false. The alert was
+skipped.
+
+That is exactly the failure this monitoring exists to prevent, so it is pinned
+here rather than left to review.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import yaml
+
+ROOT = Path(__file__).resolve().parents[1]
+WORKFLOW = ROOT / ".github" / "workflows" / "monitoring.yml"
+
+
+def _steps() -> list[dict]:
+    document = yaml.safe_load(WORKFLOW.read_text(encoding="utf-8"))
+    return document["jobs"]["verify-production-state"]["steps"]
+
+
+def test_every_measurement_disables_errexit_before_capturing_its_code() -> None:
+    for step in _steps():
+        run = str(step.get("run") or "")
+        if "GITHUB_OUTPUT" not in run or "exit_code" not in run:
+            continue
+        assert "set +e" in run, (
+            f"{step.get('name')} records an exit code, but GitHub's default shell "
+            "sets -e; without 'set +e' the step aborts before writing the output "
+            "and its alert is skipped"
+        )
+
+
+def test_every_delivery_step_runs_even_after_its_measurement_fails() -> None:
+    for step in _steps():
+        name = str(step.get("name") or "")
+        if not name.startswith("Deliver"):
+            continue
+        condition = str(step.get("if") or "")
+        assert condition.startswith("always()"), (
+            f"{name} must run after a failed measurement, otherwise a detected "
+            "outage produces no notification"
+        )
+
+
+def test_checks_do_not_block_one_another() -> None:
+    # A failing heartbeat previously skipped the release-state and capacity
+    # checks entirely, hiding whatever else was wrong.
+    ids = {"release_state", "release_runs", "capacity"}
+    for step in _steps():
+        if step.get("id") in ids:
+            assert str(step.get("if") or "").startswith("always()"), (
+                f"{step.get('name')} must not be skipped because an earlier check failed"
+            )
+
+
+def test_the_job_still_reports_unhealthy_in_its_own_colour() -> None:
+    gate = [s for s in _steps() if "Fail the job" in str(s.get("name") or "")]
+    assert gate, "the job must fail on its own when production is unhealthy"
+    run = str(gate[0]["run"])
+    # A broken alert channel must not be able to make an outage look healthy.
+    for signal in ("heartbeat", "release_state", "release_runs", "capacity"):
+        assert signal in run