fix(emergency-break): survivable self-destruct, reliable break report, ops visibility

Incident 2026-07-22: the 48h emergency break left exe + loaded DLLs orphaned
forever — a second agent instance held the images while the cleanup script
(waiting on the launcher PID only) force-deleted marker + state around the
locks AFTER removing the Scheduled Task, destroying the retry path.

Agent — cleanup script (LIFE-F7):
- wait for ALL agent processes by name (PID-reuse-proof), then force-kill
  stragglers; single-flight lock file serializes concurrent cleanup scripts
- probe-rename (GUID target) BEFORE task removal; any failure renames back
  and aborts with task + marker + files intact for a next-boot retry; the
  destructive Remove-Item-on-original fallback is gone
- script generation extracted to BuildCleanupScript, pinned by 13 tests;
  all 4 variants parser-validated + E2E-smoked (happy path, locked tree)

Agent — break report delivery:
- EmergencyReporter counts DELIVERED reports: reservation rolls back on
  failure (mark-before-send permanently suppressed a key after its first
  transport failure); optional in-reservation retries
- break path: NIC wait <=15s, 3 attempts x 15s timeout (survives Functions
  cold starts observed as >5s 503s), 75s overall bound

Backend:
- backend-materialized events (agent_emergency_break, session_timeout) now
  upsert the cross-session EventType index — they were invisible to every
  search-by-eventType surface (portal + MCP)
- new AgentEmergencyBreak ops event (Agent/Warning, once per session) so
  operators see silently-lost agents; dual-registered in OPS_EVENT_TYPES
- materialization extracted to a static seam, covered by 4 new tests

Agent 1963/1963, Backend 3198/3198, tsc clean.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/Runtime/CleanupServiceScriptTests.cs

@@ -0,0 +1,135 @@
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Runtime;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Runtime
+{
+    /// <summary>
+    /// Pins the LIFE-F7 hardening invariants of the self-destruct cleanup script
+    /// (incident 2026-07-22: a second agent instance held exe + loaded DLLs while the
+    /// old script — waiting on the launcher PID only — force-deleted around the locks
+    /// AFTER removing the Scheduled Task, orphaning the binaries with no retry path).
+    /// These are string-level pins on <see cref="CleanupService.BuildCleanupScript"/>:
+    /// weakening any of them silently re-introduces the orphaned-binaries failure mode.
+    /// </summary>
+    public class CleanupServiceScriptTests
+    {
+        private const string ProcessName = "AutopilotMonitor.Agent";
+        private const string BasePath = @"C:\ProgramData\AutopilotMonitor";
+        private const string TaskName = "AutopilotMonitorAgent";
+
+        private static string Script(bool keepLogs = false, bool reboot = false)
+            => CleanupService.BuildCleanupScript(ProcessName, BasePath, keepLogs, TaskName, reboot);
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Waits_by_process_name_not_launcher_pid(bool keepLogs)
+        {
+            var script = Script(keepLogs);
+
+            // LIFE-F7 #2: name-based wait covers marker-retry / self-update / manual instances
+            // and is immune to PID reuse. Wait-Process -Id saw exactly one PID and nothing else.
+            Assert.Contains($"Get-Process -Name '{ProcessName}'", script);
+            Assert.DoesNotContain("Wait-Process", script);
+
+            // Backstop: stragglers are force-killed — their images are deleted right after.
+            Assert.Contains("Stop-Process -Force", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Never_deletes_the_original_path(bool keepLogs)
+        {
+            var script = Script(keepLogs);
+
+            // LIFE-F7 #3: deletion happens exclusively on successfully renamed targets
+            // ($entry.Renamed). A Remove-Item against the original tree is what deleted
+            // marker + state around locked binaries in the incident.
+            Assert.DoesNotContain($"Remove-Item -Path '{BasePath}'", script);
+            Assert.Contains("Remove-Item -Path $entry.Renamed", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Task_removal_is_gated_on_probe_and_deletion_on_task_gone(bool keepLogs)
+        {
+            var script = Script(keepLogs);
+
+            // Probe-rename first; only a successful probe reaches task removal, and only a
+            // removed task reaches deletion. Both failure branches rename BACK so the next
+            // boot retries with task + marker + files intact.
+            Assert.Contains("if ($probeOk)", script);
+            Assert.Contains("if ($taskGone)", script);
+            Assert.Contains("Rename-Item -Path $entry.Renamed -NewName $entry.Original", script);
+
+            // LIFE-F6 task-removal loop + module-independent fallback stay intact.
+            Assert.Contains("Unregister-ScheduledTask -TaskName $taskName", script);
+            Assert.Contains("schtasks.exe /Delete /TN $taskName /F", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Concurrent_cleanup_scripts_are_serialized_by_lock_file(bool keepLogs)
+        {
+            var script = Script(keepLogs);
+
+            // LIFE-F7 #1: CreateNew is the atomic single-flight gate; the loser exits without
+            // touching the tree. Stale locks (>10 min) are taken over.
+            Assert.Contains("AutopilotMonitor-Cleanup.lock", script);
+            Assert.Contains("[System.IO.FileMode]::CreateNew", script);
+            Assert.Contains("$lockAge.TotalMinutes -gt 10", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Rename_targets_carry_a_guid_suffix(bool keepLogs)
+        {
+            // A fixed '.del' target collides with leftovers of earlier attempts and with a
+            // concurrent script's rename — the GUID makes every attempt's target unique.
+            Assert.Contains(".del-' + [Guid]::NewGuid().ToString('N')", Script(keepLogs));
+        }
+
+        [Fact]
+        public void KeepLogs_variant_excludes_logs_directory()
+        {
+            var script = Script(keepLogs: true);
+            Assert.Contains("-Exclude 'Logs'", script);
+            Assert.DoesNotContain($"Rename-Item -Path '{BasePath}' -NewName", script);
+        }
+
+        [Fact]
+        public void Full_variant_renames_the_whole_tree()
+        {
+            var script = Script(keepLogs: false);
+            Assert.Contains($"Rename-Item -Path '{BasePath}' -NewName $renamedPath", script);
+            Assert.DoesNotContain("-Exclude 'Logs'", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Reboot_flag_controls_shutdown_invocation(bool reboot)
+        {
+            var script = Script(keepLogs: false, reboot: reboot);
+            if (reboot)
+                Assert.Contains("shutdown.exe /r /t 10", script);
+            else
+                Assert.DoesNotContain("shutdown.exe", script);
+        }
+
+        [Theory]
+        [InlineData(false)]
+        [InlineData(true)]
+        public void Script_and_lock_are_released_in_finally(bool keepLogs)
+        {
+            var script = Script(keepLogs);
+            Assert.Contains("finally", script);
+            Assert.Contains("$lockStream.Dispose()", script);
+            Assert.Contains("Remove-Item -Path $scriptPath -Force", script);
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core.Tests/Monitoring/Transport/EmergencyReporterTests.cs

@@ -0,0 +1,143 @@
+using System;
+using System.Collections.Generic;
+using System.Net.Http;
+using System.Threading.Tasks;
+using AutopilotMonitor.Agent.V2.Core.Monitoring.Transport;
+using AutopilotMonitor.Shared.Models;
+using Xunit;
+
+namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Transport
+{
+    /// <summary>
+    /// Pins the anti-flood semantics of <see cref="EmergencyReporter.TrySendAsync"/> after the
+    /// delivery-hardening change (2026-07-23): the once-per-session dedup and the per-session
+    /// budget count DELIVERED reports. The previous mark-before-send permanently suppressed an
+    /// error key after its first transport failure — fatal for single-shot callers like the 48h
+    /// emergency break, whose process exits right after the attempt.
+    /// </summary>
+    public class EmergencyReporterTests
+    {
+        private sealed class FakeApiClient : BackendApiClient
+        {
+            public readonly List<TimeSpan?> Timeouts = new List<TimeSpan?>();
+            public readonly Queue<bool> Results = new Queue<bool>();
+
+            public FakeApiClient() : base(
+                httpClient: new HttpClient(),
+                baseUrl: "http://localhost",
+                manufacturer: string.Empty,
+                model: string.Empty,
+                serialNumber: string.Empty,
+                useBootstrapTokenAuth: false,
+                bootstrapToken: null,
+                agentVersion: "0.0.0",
+                logger: null)
+            {
+            }
+
+            public int Calls => Timeouts.Count;
+
+            public override Task<bool> ReportAgentErrorAsync(AgentErrorReport report, TimeSpan? timeout = null)
+            {
+                Timeouts.Add(timeout);
+                return Task.FromResult(Results.Count > 0 && Results.Dequeue());
+            }
+        }
+
+        private static EmergencyReporter Reporter(FakeApiClient client)
+            => new EmergencyReporter(client, "session", "tenant", "0.0.0", logger: null);
+
+        [Fact]
+        public async Task Failed_send_rolls_back_reservation_so_a_later_call_retries()
+        {
+            var client = new FakeApiClient();
+            client.Results.Enqueue(false); // first call fails
+            client.Results.Enqueue(true);  // second call succeeds
+            var reporter = Reporter(client);
+
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+
+            // Old behaviour: second call suppressed as "already reported" ⇒ 1 call.
+            Assert.Equal(2, client.Calls);
+        }
+
+        [Fact]
+        public async Task Delivered_send_is_deduplicated_per_key()
+        {
+            var client = new FakeApiClient();
+            client.Results.Enqueue(true);
+            var reporter = Reporter(client);
+
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+
+            Assert.Equal(1, client.Calls);
+        }
+
+        [Fact]
+        public async Task Retries_stay_inside_one_reservation_and_stop_on_success()
+        {
+            var client = new FakeApiClient();
+            client.Results.Enqueue(false);
+            client.Results.Enqueue(true); // second attempt delivers — third must not fire
+            var reporter = Reporter(client);
+
+            await reporter.TrySendAsync(
+                AgentErrorType.SessionAgeEmergencyBreak, "m",
+                attempts: 3, perAttemptTimeout: TimeSpan.FromMilliseconds(50), retryDelay: TimeSpan.Zero);
+
+            Assert.Equal(2, client.Calls);
+
+            // Delivered ⇒ dedup holds for the session.
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+            Assert.Equal(2, client.Calls);
+        }
+
+        [Fact]
+        public async Task Per_attempt_timeout_is_forwarded_to_the_api_client()
+        {
+            var client = new FakeApiClient();
+            client.Results.Enqueue(true);
+            var reporter = Reporter(client);
+
+            var timeout = TimeSpan.FromSeconds(15);
+            await reporter.TrySendAsync(
+                AgentErrorType.SessionAgeEmergencyBreak, "m",
+                attempts: 1, perAttemptTimeout: timeout);
+
+            Assert.Equal(timeout, Assert.Single(client.Timeouts));
+        }
+
+        [Fact]
+        public async Task All_attempts_failing_rolls_back_after_exhausting_attempts()
+        {
+            var client = new FakeApiClient(); // queue empty ⇒ every attempt fails
+            var reporter = Reporter(client);
+
+            await reporter.TrySendAsync(
+                AgentErrorType.SessionAgeEmergencyBreak, "m",
+                attempts: 3, retryDelay: TimeSpan.Zero);
+            Assert.Equal(3, client.Calls);
+
+            // Rolled back ⇒ a later call may try again (budget + dedup + cooldown restored).
+            client.Results.Enqueue(true);
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+            Assert.Equal(4, client.Calls);
+        }
+
+        [Fact]
+        public async Task Different_keys_do_not_suppress_each_other_after_rollback()
+        {
+            var client = new FakeApiClient();
+            client.Results.Enqueue(false); // break report fails → rollback (incl. cooldown restore)
+            client.Results.Enqueue(true);  // unrelated report must still go out
+            var reporter = Reporter(client);
+
+            await reporter.TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, "m");
+            await reporter.TrySendAsync(AgentErrorType.ConfigFetchFailed, "m");
+
+            Assert.Equal(2, client.Calls);
+        }
+    }
+}

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Runtime/CleanupService.cs

@@ -31,84 +31,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Runtime
             {
                 _logger.Info($"Executing FULL SELF-DESTRUCT (Scheduled Task + Files{(_configuration.RebootOnComplete ? " + Reboot" : "")})");
 
-                var currentProcessId = Process.GetCurrentProcess().Id;
+                var currentProcess = Process.GetCurrentProcess();
                 var agentBasePath = Path.GetFullPath(Path.Combine(
                     Environment.GetFolderPath(Environment.SpecialFolder.CommonApplicationData),
                     "AutopilotMonitor"
                 ));
 
-                var logDir = Path.GetFullPath(Path.Combine(agentBasePath, "Logs"));
-                var keepLogs = _configuration.KeepLogFile;
-
-                // Create a self-deleting PowerShell script
-                var cleanupScript = $@"
-$scriptPath = $MyInvocation.MyCommand.Path
-
-# Wait for agent process to exit
-Start-Sleep -Seconds 2
-Wait-Process -Id {currentProcessId} -ErrorAction SilentlyContinue -Timeout 30
-Start-Sleep -Seconds 1
-
-# Remove Scheduled Task — verify removal BEFORE deleting files (LIFE-F6). An orphan task firing
-# against a deleted exe spams Task Scheduler errors every boot forever and, once the files (incl.
-# the enrollment-complete marker) are gone, has no cleanup-retry path. So confirm the task is gone
-# first; if it cannot be removed, SKIP the file deletion and leave everything intact for a clean
-# cleanup retry on the next boot (the marker survives, so bootstrap re-runs cleanup).
-$taskName = '{_configuration.ScheduledTaskName}'
-$taskGone = $false
-for ($i = 1; $i -le 5; $i++) {{
-    try {{ Stop-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue }} catch {{ }}
-    try {{ Unregister-ScheduledTask -TaskName $taskName -Confirm:$false -ErrorAction SilentlyContinue }} catch {{ }}
-    $still = $null
-    try {{ $still = Get-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue }} catch {{ }}
-    if (-not $still) {{ $taskGone = $true; break }}
-    Start-Sleep -Seconds 1
-}}
-if (-not $taskGone) {{
-    # Fallback: schtasks.exe /Delete does not depend on the ScheduledTasks PowerShell module.
-    try {{ & schtasks.exe /Delete /TN $taskName /F 2>$null | Out-Null }} catch {{ }}
-    try {{ $taskGone = -not (Get-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue) }} catch {{ }}
-}}
-
-{(keepLogs ? $@"
-if ($taskGone) {{
-    # Delete everything except Logs directory
-    Get-ChildItem -Path '{agentBasePath}' -Exclude 'Logs' | ForEach-Object {{
-        $dest = $_.FullName + '.del'
-        try {{ Rename-Item -Path $_.FullName -NewName $dest -Force -ErrorAction SilentlyContinue }} catch {{ }}
-        Remove-Item -Path $dest -Recurse -Force -ErrorAction SilentlyContinue
-    }}
-}}
-" : $@"
-if ($taskGone) {{
-    # Rename folder then delete. Retry a few times to let the OS release handles.
-    $renamedPath = '{agentBasePath}.del'
-    $renamed = $false
-    for ($i = 1; $i -le 10; $i++) {{
-        try {{
-            Rename-Item -Path '{agentBasePath}' -NewName $renamedPath -Force -ErrorAction Stop
-            $renamed = $true
-            break
-        }} catch {{
-            Start-Sleep -Seconds 2
-        }}
-    }}
-    if ($renamed) {{
-        Remove-Item -Path $renamedPath -Recurse -Force -ErrorAction SilentlyContinue
-    }} else {{
-        Remove-Item -Path '{agentBasePath}' -Recurse -Force -ErrorAction SilentlyContinue
-    }}
-}}
-")}
-{(_configuration.RebootOnComplete ? @"
-# shutdown.exe (NOT Restart-Computer): Restart-Computer has no -Comment parameter and -Delay is
-# only honoured with -Wait, so the previous invocation failed parameter binding and the device
-# silently never rebooted. shutdown.exe /r /t 10 mirrors the standalone reboot path and reboots
-# after a 10 s delay (lets this script finish + the agent exit). Review LIFE-F2.
-shutdown.exe /r /t 10 /c 'Autopilot enrollment completed - Autopilot Monitor is rebooting'
-" : "")}
-Remove-Item -Path $scriptPath -Force -ErrorAction SilentlyContinue
-";
+                var cleanupScript = BuildCleanupScript(
+                    agentProcessName: currentProcess.ProcessName,
+                    agentBasePath: agentBasePath,
+                    keepLogs: _configuration.KeepLogFile,
+                    scheduledTaskName: _configuration.ScheduledTaskName,
+                    rebootOnComplete: _configuration.RebootOnComplete);
 
                 // Write cleanup script to temp location (outside of agent folder)
                 var tempScriptPath = Path.Combine(Path.GetTempPath(), $"AutopilotMonitor-Cleanup-{Guid.NewGuid()}.ps1");
@@ -143,5 +77,162 @@ Remove-Item -Path $scriptPath -Force -ErrorAction SilentlyContinue
             }
         }
 
+        /// <summary>
+        /// Builds the self-deleting cleanup PowerShell script. Extracted (internal static, pure)
+        /// so tests can pin the hardened invariants without launching processes.
+        /// <para>
+        /// LIFE-F7 hardening (incident 2026-07-22 — exe + loaded DLLs orphaned forever):
+        /// </para>
+        /// <list type="number">
+        ///   <item><b>Single-flight lock</b> — the enrollment-complete marker is written BEFORE
+        ///     cleanup launches, so any agent start during an in-flight cleanup spawns a SECOND
+        ///     cleanup script. Two unsynchronized scripts interleave their rename/delete phases
+        ///     and leave partial trees behind. A CreateNew lock file in the SYSTEM temp dir
+        ///     serializes them; a stale lock (&gt;10 min) is taken over.</item>
+        ///   <item><b>Wait by process NAME, not launcher PID</b> — marker-retry instances,
+        ///     self-update restarts and manual runs all load the same images but have PIDs the
+        ///     old <c>Wait-Process -Id</c> never saw (and an already-freed PID can be reused by
+        ///     an unrelated boot-time process). After the deadline any straggler is force-killed:
+        ///     its images are about to be deleted anyway.</item>
+        ///   <item><b>Probe-rename BEFORE task removal, never delete the original path</b> — the
+        ///     rename is the atomic lock probe. If it fails, something still holds the tree: abort
+        ///     with task + files + marker intact so the next boot retries (file-side symmetry to
+        ///     the LIFE-F6 task-side rule). If the task cannot be removed afterwards, rename BACK
+        ///     and abort for the same reason. Only a fully renamed tree with the task gone is
+        ///     deleted — the old <c>Remove-Item</c>-on-original fallback deleted marker + state
+        ///     around locked binaries AFTER the task was gone, converting a transient lock into a
+        ///     permanently orphaned exe.</item>
+        /// </list>
+        /// </summary>
+        internal static string BuildCleanupScript(
+            string agentProcessName,
+            string agentBasePath,
+            bool keepLogs,
+            string scheduledTaskName,
+            bool rebootOnComplete)
+        {
+            return $@"
+$scriptPath = $MyInvocation.MyCommand.Path
+
+# Single-flight lock (LIFE-F7 #1): serialize concurrent cleanup scripts. CreateNew is
+# atomic; the loser backs off and lets the winner own the deletion.
+$lockPath = Join-Path ([System.IO.Path]::GetTempPath()) 'AutopilotMonitor-Cleanup.lock'
+$lockStream = $null
+try {{
+    if (Test-Path $lockPath) {{
+        $lockAge = (Get-Date) - (Get-Item $lockPath -ErrorAction Stop).LastWriteTime
+        if ($lockAge.TotalMinutes -gt 10) {{ Remove-Item -Path $lockPath -Force -ErrorAction SilentlyContinue }}
+    }}
+    $lockStream = [System.IO.File]::Open($lockPath, [System.IO.FileMode]::CreateNew, [System.IO.FileAccess]::Write, [System.IO.FileShare]::None)
+}} catch {{
+    Remove-Item -Path $scriptPath -Force -ErrorAction SilentlyContinue
+    exit 0
+}}
+
+try {{
+    # Wait for EVERY agent process to exit, by NAME (LIFE-F7 #2) — covers marker-retry
+    # instances, self-update restarts and manual runs, and is immune to PID reuse.
+    Start-Sleep -Seconds 2
+    $deadline = (Get-Date).AddSeconds(60)
+    while ((Get-Date) -lt $deadline) {{
+        if (-not (Get-Process -Name '{agentProcessName}' -ErrorAction SilentlyContinue)) {{ break }}
+        Start-Sleep -Seconds 2
+    }}
+    # Backstop: force-kill stragglers — their images are about to be deleted anyway.
+    Get-Process -Name '{agentProcessName}' -ErrorAction SilentlyContinue | Stop-Process -Force -ErrorAction SilentlyContinue
+    Start-Sleep -Seconds 1
+
+{(keepLogs ? $@"
+    # Probe-rename all children except Logs (LIFE-F7 #3). Any failure ⇒ rename back and
+    # abort with the Scheduled Task intact so the next boot retries cleanly.
+    $renamed = @()
+    $probeOk = $true
+    foreach ($item in @(Get-ChildItem -Path '{agentBasePath}' -Exclude 'Logs' -ErrorAction SilentlyContinue)) {{
+        $dest = $item.FullName + '.del-' + [Guid]::NewGuid().ToString('N')
+        $itemOk = $false
+        for ($i = 1; $i -le 10; $i++) {{
+            try {{
+                Rename-Item -Path $item.FullName -NewName $dest -Force -ErrorAction Stop
+                $itemOk = $true
+                break
+            }} catch {{
+                Start-Sleep -Seconds 2
+            }}
+        }}
+        if ($itemOk) {{ $renamed += @{{ Original = $item.FullName; Renamed = $dest }} }}
+        else {{ $probeOk = $false; break }}
+    }}
+" : $@"
+    # Probe-rename the whole tree (LIFE-F7 #3). Failure ⇒ abort with task + files + marker
+    # intact so the next boot retries cleanly. GUID suffix avoids collisions with leftovers
+    # of earlier attempts.
+    $renamedPath = '{agentBasePath}.del-' + [Guid]::NewGuid().ToString('N')
+    $probeOk = $false
+    for ($i = 1; $i -le 10; $i++) {{
+        try {{
+            Rename-Item -Path '{agentBasePath}' -NewName $renamedPath -Force -ErrorAction Stop
+            $probeOk = $true
+            break
+        }} catch {{
+            Start-Sleep -Seconds 2
+        }}
+    }}
+    $renamed = @(@{{ Original = '{agentBasePath}'; Renamed = $renamedPath }})
+")}
+    if ($probeOk) {{
+        # Remove Scheduled Task — verify removal BEFORE deleting files (LIFE-F6). An orphan
+        # task firing against a deleted exe spams Task Scheduler errors every boot forever
+        # and, once the files (incl. the enrollment-complete marker) are gone, has no
+        # cleanup-retry path.
+        $taskName = '{scheduledTaskName}'
+        $taskGone = $false
+        for ($i = 1; $i -le 5; $i++) {{
+            try {{ Stop-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue }} catch {{ }}
+            try {{ Unregister-ScheduledTask -TaskName $taskName -Confirm:$false -ErrorAction SilentlyContinue }} catch {{ }}
+            $still = $null
+            try {{ $still = Get-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue }} catch {{ }}
+            if (-not $still) {{ $taskGone = $true; break }}
+            Start-Sleep -Seconds 1
+        }}
+        if (-not $taskGone) {{
+            # Fallback: schtasks.exe /Delete does not depend on the ScheduledTasks PowerShell module.
+            try {{ & schtasks.exe /Delete /TN $taskName /F 2>$null | Out-Null }} catch {{ }}
+            try {{ $taskGone = -not (Get-ScheduledTask -TaskName $taskName -ErrorAction SilentlyContinue) }} catch {{ }}
+        }}
+
+        if ($taskGone) {{
+            # Task gone + tree renamed ⇒ safe to delete. Never touches the original path.
+            foreach ($entry in $renamed) {{
+                Remove-Item -Path $entry.Renamed -Recurse -Force -ErrorAction SilentlyContinue
+            }}
+        }} else {{
+            # Task survived ⇒ rename back and leave everything for the next-boot retry
+            # (the marker survives, so bootstrap re-runs cleanup).
+            foreach ($entry in $renamed) {{
+                try {{ Rename-Item -Path $entry.Renamed -NewName $entry.Original -Force -ErrorAction SilentlyContinue }} catch {{ }}
+            }}
+        }}
+    }} else {{
+        # Probe failed ⇒ something still holds the tree. Rename back whatever succeeded and
+        # abort WITHOUT deleting anything — task + marker stay, next boot retries.
+        foreach ($entry in $renamed) {{
+            try {{ Rename-Item -Path $entry.Renamed -NewName $entry.Original -Force -ErrorAction SilentlyContinue }} catch {{ }}
+        }}
+    }}
+{(rebootOnComplete ? @"
+    # shutdown.exe (NOT Restart-Computer): Restart-Computer has no -Comment parameter and -Delay is
+    # only honoured with -Wait, so the previous invocation failed parameter binding and the device
+    # silently never rebooted. shutdown.exe /r /t 10 mirrors the standalone reboot path and reboots
+    # after a 10 s delay (lets this script finish + the agent exit). Review LIFE-F2.
+    shutdown.exe /r /t 10 /c 'Autopilot enrollment completed - Autopilot Monitor is rebooting'
+" : "")}
+}} finally {{
+    if ($lockStream) {{ $lockStream.Dispose() }}
+    Remove-Item -Path $lockPath -Force -ErrorAction SilentlyContinue
+    Remove-Item -Path $scriptPath -Force -ErrorAction SilentlyContinue
+}}
+";
+        }
+
     }
 }

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Transport/BackendApiClient.cs

@@ -169,9 +169,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
         /// <summary>
         /// Sends a critical error report to the emergency channel endpoint.
         /// Fire-and-forget: swallows all exceptions so a failure here never cascades.
-        /// Uses a 5-second per-request timeout so it cannot block the upload loop.
+        /// Default 5-second per-request timeout so it cannot block the upload loop; the
+        /// emergency-break path passes a longer <paramref name="timeout"/> because its first
+        /// request of a freshly booted process pays cold TLS + DNS + a possible backend
+        /// cold start (observed 503s take well over 5 s).
+        /// Returns <c>true</c> when the HTTP round-trip completed (the endpoint always
+        /// answers 200), <c>false</c> on timeout/transport failure — the caller-side
+        /// anti-flood layer uses this to decide whether the report counts as delivered.
+        /// Virtual for tests.
         /// </summary>
-        public async Task ReportAgentErrorAsync(AgentErrorReport report)
+        public virtual async Task<bool> ReportAgentErrorAsync(AgentErrorReport report, TimeSpan? timeout = null)
         {
             try
             {
@@ -190,16 +197,18 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
                     httpRequest.Headers.Add("X-Tenant-Id", report.TenantId);
                     AddSecurityHeaders(httpRequest);
 
-                    using (var cts = new CancellationTokenSource(TimeSpan.FromSeconds(5)))
+                    using (var cts = new CancellationTokenSource(timeout ?? TimeSpan.FromSeconds(5)))
                     using (var response = await _httpClient.SendAsync(httpRequest, cts.Token).ConfigureAwait(false))
                     {
                         // Response status is deliberately ignored — endpoint always returns 200.
+                        return true;
                     }
                 }
             }
             catch (Exception ex)
             {
                 _logger?.Debug($"ReportAgentErrorAsync: emergency channel failed: {ex.Message}");
+                return false;
             }
         }
 

src/Agent/AutopilotMonitor.Agent.V2.Core/Monitoring/Transport/EmergencyReporter.cs

@@ -57,19 +57,33 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
         /// (intended to be called as fire-and-forget via <c>_ = reporter.TrySendAsync(...)</c>).
         ///
         /// All anti-flood checks are applied synchronously before the async HTTP call,
-        /// so repeated calls are cheap when suppressed.
+        /// so repeated calls are cheap when suppressed. The anti-flood reservation is
+        /// taken up front (so concurrent same-key calls cannot double-send) but ROLLED
+        /// BACK when every attempt fails: the budget and the once-per-session dedup are
+        /// meant to count DELIVERED reports — the previous mark-before-send made the very
+        /// first transport failure permanently suppress that error key for the session
+        /// (found via the 48h emergency-break report, which is a single shot at boot).
+        ///
+        /// <paramref name="attempts"/>/<paramref name="perAttemptTimeout"/>/<paramref name="retryDelay"/>
+        /// exist for callers whose process is about to exit and cannot come back later
+        /// (emergency break): retries stay INSIDE the one reservation, so they never
+        /// multiply against the per-session budget.
         /// </summary>
         public virtual async Task TrySendAsync(
             AgentErrorType errorType,
             string message,
             int? httpStatusCode = null,
-            long? sequenceNumber = null)
+            long? sequenceNumber = null,
+            int attempts = 1,
+            TimeSpan? perAttemptTimeout = null,
+            TimeSpan? retryDelay = null)
         {
             // Deduplicate by error type + status code: same failure category sent only once per session.
             // All anti-flood checks are protected by a lock to prevent race conditions when
             // TrySendAsync is called concurrently (fire-and-forget from multiple threads).
             var key = $"{errorType}:{httpStatusCode}";
             int currentReportCount;
+            DateTime? previousLastReportTime;
 
             lock (_antiFloodLock)
             {
@@ -92,6 +106,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
                     return;
                 }
 
+                previousLastReportTime = _lastReportTime;
                 _reportedKeys.Add(key);
                 _reportCount++;
                 _lastReportTime = DateTime.UtcNow;
@@ -113,15 +128,40 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Transport
                 Timestamp = DateTime.UtcNow
             };
 
+            var delivered = false;
+            var totalAttempts = attempts < 1 ? 1 : attempts;
             try
             {
-                await _apiClient.ReportAgentErrorAsync(report);
+                for (var attempt = 1; attempt <= totalAttempts && !delivered; attempt++)
+                {
+                    if (attempt > 1 && retryDelay.HasValue)
+                    {
+                        await Task.Delay(retryDelay.Value).ConfigureAwait(false);
+                    }
+
+                    delivered = await _apiClient.ReportAgentErrorAsync(report, perAttemptTimeout).ConfigureAwait(false);
+                    if (!delivered && totalAttempts > 1)
+                    {
+                        _logger?.Debug($"[EmergencyChannel] Attempt {attempt}/{totalAttempts} failed: {key}");
+                    }
+                }
             }
             catch (Exception ex)
             {
                 // Swallow — emergency channel must never cascade failures into the caller
                 _logger?.Debug($"[EmergencyChannel] Failed to send report: {ex.Message}");
             }
+
+            if (!delivered)
+            {
+                lock (_antiFloodLock)
+                {
+                    _reportedKeys.Remove(key);
+                    _reportCount--;
+                    _lastReportTime = previousLastReportTime;
+                }
+                _logger?.Debug($"[EmergencyChannel] Report not delivered — reservation rolled back: {key}");
+            }
         }
     }
 }

src/Agent/AutopilotMonitor.Agent.V2/Runtime/EmergencyBreakReporter.cs

@@ -1,4 +1,6 @@
 using System;
+using System.Net.NetworkInformation;
+using System.Threading;
 using AutopilotMonitor.Agent.V2.Core.Configuration;
 using AutopilotMonitor.Agent.V2.Core.Logging;
 using AutopilotMonitor.Shared.Models;
@@ -15,28 +17,78 @@ namespace AutopilotMonitor.Agent.V2.Runtime
     /// classifier terminalize the session instead of waiting out the grace. Fully swallowed — a send
     /// failure (no network, no cert) must never delay the cleanup/exit the break exists to perform.
     /// See tasks/enrollment-status-reclassification.md.
+    /// <para>
+    /// Delivery hardening (2026-07-23): this is the FIRST request of a freshly booted process and
+    /// the process never comes back — a miss is gone forever. The old single shot with a 5 s cap
+    /// died on boot-time realities (NIC still associating, cold TLS/DNS, backend cold starts
+    /// observed as &gt;5 s 503s). Now: wait briefly for the NIC, then up to
+    /// <see cref="SendAttempts"/> attempts with a per-attempt timeout that survives a Functions
+    /// cold start. The session is 48 h dead — spending up to ~75 s here is irrelevant, losing the
+    /// only signal that the agent gave up is not.
+    /// </para>
     /// </summary>
     internal static class EmergencyBreakReporter
     {
+        private static readonly TimeSpan NetworkWaitMax = TimeSpan.FromSeconds(15);
+        private const int SendAttempts = 3;
+        private static readonly TimeSpan PerAttemptTimeout = TimeSpan.FromSeconds(15);
+        private static readonly TimeSpan RetryDelay = TimeSpan.FromSeconds(5);
+        // Upper bound for the blocking wait: NIC wait + 3×15s attempts + 2×5s delays + slack.
+        private static readonly TimeSpan OverallBudget = TimeSpan.FromSeconds(75);
+
         public static void TrySend(AgentConfiguration agentConfig, string agentVersion, AgentLogger logger)
         {
             try
             {
+                WaitForNetwork(logger);
+
                 var auth = BackendClientFactory.BuildAuthClients(agentConfig, agentVersion, logger);
                 var message =
                     $"Agent absolute session-age emergency break fired (cap {agentConfig.AbsoluteMaxSessionHours}h) — cleaning up and exiting.";
 
-                // The process is about to exit, so block briefly on the otherwise fire-and-forget send
-                // rather than let it be abandoned mid-flight. TrySendAsync swallows its own HTTP errors;
-                // the auth bundle's HttpClients are intentionally left for process teardown.
+                // The process is about to exit, so block on the otherwise fire-and-forget send
+                // rather than let it be abandoned mid-flight. TrySendAsync swallows its own HTTP
+                // errors and keeps the retries inside one anti-flood reservation; the auth
+                // bundle's HttpClients are intentionally left for process teardown.
                 auth.EmergencyReporter
-                    .TrySendAsync(AgentErrorType.SessionAgeEmergencyBreak, message)
-                    .Wait(TimeSpan.FromSeconds(5));
+                    .TrySendAsync(
+                        AgentErrorType.SessionAgeEmergencyBreak,
+                        message,
+                        attempts: SendAttempts,
+                        perAttemptTimeout: PerAttemptTimeout,
+                        retryDelay: RetryDelay)
+                    .Wait(OverallBudget);
             }
             catch (Exception ex)
             {
                 logger?.Debug($"Emergency-break report send failed (best-effort): {ex.Message}");
             }
         }
+
+        /// <summary>
+        /// Boot-time NIC grace: the break fires during bootstrap, often seconds after boot while
+        /// Wi-Fi is still associating. Polls the cheap link-level signal — it cannot prove backend
+        /// reachability (the send attempts handle that), it only avoids burning the first attempts
+        /// into a known-dead link. Best-effort: probe errors abort the wait, never the send.
+        /// </summary>
+        private static void WaitForNetwork(AgentLogger logger)
+        {
+            try
+            {
+                if (NetworkInterface.GetIsNetworkAvailable()) return;
+
+                logger?.Info($"Emergency-break report: no network link yet — waiting up to {NetworkWaitMax.TotalSeconds:F0}s.");
+                var deadline = DateTime.UtcNow + NetworkWaitMax;
+                while (DateTime.UtcNow < deadline)
+                {
+                    Thread.Sleep(TimeSpan.FromSeconds(1));
+                    if (NetworkInterface.GetIsNetworkAvailable()) return;
+                }
+            }
+            catch (Exception ex)
+            {
+                logger?.Debug($"Emergency-break report: network probe failed ({ex.Message}) — sending anyway.");
+            }
+        }
     }
 }

src/Backend/AutopilotMonitor.Functions.Tests/ReportAgentErrorFunctionTests.cs

@@ -1,7 +1,15 @@
 using System;
 using System.Collections.Generic;
+using System.Net.Http;
+using System.Threading.Tasks;
 using AutopilotMonitor.Functions.Functions.Ingest;
+using AutopilotMonitor.Functions.Services;
+using AutopilotMonitor.Functions.Services.Notifications;
+using AutopilotMonitor.Shared.DataAccess;
 using AutopilotMonitor.Shared.Models;
+using Microsoft.Extensions.Caching.Memory;
+using Microsoft.Extensions.Logging.Abstractions;
+using Moq;
 
 namespace AutopilotMonitor.Functions.Tests;
 
@@ -80,4 +88,116 @@ public class ReportAgentErrorFunctionTests
         var evt = ReportAgentErrorFunction.BuildAgentEmergencyBreakEvent(Report(Now, message: ""), TenantId, null!, Now);
         Assert.Contains("emergency break", evt.Message, StringComparison.OrdinalIgnoreCase);
     }
+
+    // =========================================================================
+    // MaterializeEmergencyBreakArtifactsAsync — the artifact set per break report
+    // (2026-07-23 hardening): timeline event + cross-session EventType index row +
+    // AgentEmergencyBreak ops event, once per session. The index upsert is load-bearing:
+    // StoreEventsBatchAsync alone leaves the event invisible to every search-by-eventType
+    // surface (found in the 2026-07-22 incident analysis).
+    // =========================================================================
+
+    private sealed class Harness
+    {
+        public readonly Mock<ISessionRepository> SessionRepo = new();
+        public readonly List<OpsEventEntry> OpsEvents = new();
+        public readonly OpsEventService OpsService;
+
+        public Harness(List<EnrollmentEvent>? existingEvents = null)
+        {
+            SessionRepo.Setup(r => r.GetSessionEventsAsync(TenantId, SessionId, It.IsAny<int>()))
+                .ReturnsAsync(existingEvents ?? new List<EnrollmentEvent>());
+            SessionRepo.Setup(r => r.StoreEventsBatchAsync(It.IsAny<List<EnrollmentEvent>>()))
+                .ReturnsAsync((List<EnrollmentEvent> e) => e);
+            SessionRepo.Setup(r => r.UpsertEventTypeIndexBatchAsync(
+                    It.IsAny<string>(), It.IsAny<string>(), It.IsAny<IEnumerable<EnrollmentEvent>>()))
+                .Returns(Task.CompletedTask);
+
+            var opsRepo = new Mock<IOpsEventRepository>();
+            opsRepo.Setup(r => r.SaveOpsEventAsync(It.IsAny<OpsEventEntry>()))
+                .Callback<OpsEventEntry>(e => { lock (OpsEvents) OpsEvents.Add(e); })
+                .Returns(Task.CompletedTask);
+
+            var adminConfig = new Mock<AdminConfigurationService>(
+                Mock.Of<IConfigRepository>(), NullLogger<AdminConfigurationService>.Instance,
+                new MemoryCache(new MemoryCacheOptions()));
+            var alertDispatch = new OpsAlertDispatchService(
+                adminConfig.Object,
+                new TelegramNotificationService(new HttpClient(), Mock.Of<IConfigRepository>(),
+                    NullLogger<TelegramNotificationService>.Instance),
+                new WebhookNotificationService(new HttpClient(),
+                    NullLogger<WebhookNotificationService>.Instance),
+                NullLogger<OpsAlertDispatchService>.Instance);
+            OpsService = new OpsEventService(opsRepo.Object, NullLogger<OpsEventService>.Instance, alertDispatch);
+        }
+
+        public Task RunAsync(AgentErrorReport report)
+            => ReportAgentErrorFunction.MaterializeEmergencyBreakArtifactsAsync(
+                report, TenantId, SessionRepo.Object, OpsService, NullLogger<ReportAgentErrorFunction>.Instance);
+    }
+
+    [Fact]
+    public async Task First_break_report_materializes_event_index_row_and_ops_event()
+    {
+        var h = new Harness();
+
+        await h.RunAsync(Report(Now));
+
+        h.SessionRepo.Verify(r => r.StoreEventsBatchAsync(
+            It.Is<List<EnrollmentEvent>>(e => e.Count == 1
+                && e[0].EventType == AutopilotMonitor.Shared.Constants.EventTypes.AgentEmergencyBreak)), Times.Once);
+        h.SessionRepo.Verify(r => r.UpsertEventTypeIndexBatchAsync(
+            TenantId, SessionId, It.IsAny<IEnumerable<EnrollmentEvent>>()), Times.Once);
+
+        var ops = Assert.Single(h.OpsEvents);
+        Assert.Equal("AgentEmergencyBreak", ops.EventType);
+        Assert.Equal(TenantId, ops.TenantId);
+    }
+
+    [Fact]
+    public async Task Already_materialized_session_produces_no_second_artifact_set()
+    {
+        var h = new Harness(new List<EnrollmentEvent>
+        {
+            new() { EventType = AutopilotMonitor.Shared.Constants.EventTypes.AgentEmergencyBreak, Sequence = 45 },
+        });
+
+        await h.RunAsync(Report(Now));
+
+        h.SessionRepo.Verify(r => r.StoreEventsBatchAsync(It.IsAny<List<EnrollmentEvent>>()), Times.Never);
+        h.SessionRepo.Verify(r => r.UpsertEventTypeIndexBatchAsync(
+            It.IsAny<string>(), It.IsAny<string>(), It.IsAny<IEnumerable<EnrollmentEvent>>()), Times.Never);
+        Assert.Empty(h.OpsEvents);
+    }
+
+    [Fact]
+    public async Task Non_break_error_types_and_missing_session_id_are_ignored()
+    {
+        var h = new Harness();
+
+        var other = Report(Now);
+        other.ErrorType = AgentErrorType.ConfigFetchFailed;
+        await h.RunAsync(other);
+
+        var noSession = Report(Now);
+        noSession.SessionId = null!;
+        await h.RunAsync(noSession);
+
+        h.SessionRepo.Verify(r => r.GetSessionEventsAsync(
+            It.IsAny<string>(), It.IsAny<string>(), It.IsAny<int>()), Times.Never);
+        Assert.Empty(h.OpsEvents);
+    }
+
+    [Fact]
+    public async Task Storage_failure_is_swallowed_and_never_throws()
+    {
+        var h = new Harness();
+        h.SessionRepo.Setup(r => r.StoreEventsBatchAsync(It.IsAny<List<EnrollmentEvent>>()))
+            .ThrowsAsync(new InvalidOperationException("storage down"));
+
+        // Must not throw — the always-200 emergency channel must never turn into a retry loop.
+        await h.RunAsync(Report(Now));
+
+        Assert.Empty(h.OpsEvents);
+    }
 }

src/Backend/AutopilotMonitor.Functions/Functions/Ingest/ReportAgentErrorFunction.cs

@@ -37,6 +37,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
         private readonly TelemetryClient _telemetryClient;
         private readonly BootstrapSessionService _bootstrapSessionService;
         private readonly ISessionRepository _sessionRepo;
+        private readonly OpsEventService _opsEventService;
 
         public ReportAgentErrorFunction(
             ILogger<ReportAgentErrorFunction> logger,
@@ -48,7 +49,8 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             DeviceAssociationValidator deviceAssociationValidator,
             TelemetryClient telemetryClient,
             BootstrapSessionService bootstrapSessionService,
-            ISessionRepository sessionRepo)
+            ISessionRepository sessionRepo,
+            OpsEventService opsEventService)
         {
             _logger = logger;
             _configService = configService;
@@ -60,6 +62,7 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
             _telemetryClient = telemetryClient;
             _bootstrapSessionService = bootstrapSessionService;
             _sessionRepo = sessionRepo;
+            _opsEventService = opsEventService;
         }
 
         [Function("ReportAgentError")]
@@ -169,35 +172,71 @@ namespace AutopilotMonitor.Functions.Functions.Ingest
                 ["AgentTimestamp"] = report.Timestamp.ToString("O"),
             });
 
-            // Materialize the agent's silent 48h emergency break as a timeline event so it shows in the
-            // session and the timeout classifier can see it (tasks/enrollment-status-reclassification.md).
-            // Best-effort — a failure here must never turn the always-200 emergency channel into a retry loop.
-            if (report.ErrorType == AgentErrorType.SessionAgeEmergencyBreak
-                && !string.IsNullOrEmpty(report.SessionId))
-            {
-                try
-                {
-                    var existing = await _sessionRepo.GetSessionEventsAsync(tenantId, report.SessionId, maxResults: 1000);
-                    // Idempotency: the agent's emergency channel can send up to a few reports per session;
-                    // only ever materialize one timeline event.
-                    var alreadyMaterialized = existing.Any(e =>
-                        string.Equals(e.EventType, Constants.EventTypes.AgentEmergencyBreak, StringComparison.OrdinalIgnoreCase));
-                    if (!alreadyMaterialized)
-                    {
-                        var evt = BuildAgentEmergencyBreakEvent(report, tenantId, existing, DateTime.UtcNow);
-                        await _sessionRepo.StoreEventsBatchAsync(new List<EnrollmentEvent> { evt });
-                    }
-                }
-                catch (Exception ex)
-                {
-                    _logger.LogWarning(ex,
-                        "ReportAgentError: failed to materialize agent_emergency_break event for session {SessionId}", report.SessionId);
-                }
-            }
+            await MaterializeEmergencyBreakArtifactsAsync(report, tenantId, _sessionRepo, _opsEventService, _logger);
 
             return req.CreateResponse(HttpStatusCode.OK);
         }
 
+        /// <summary>
+        /// Materializes the agent's silent 48h emergency break as (1) a timeline event so it shows
+        /// in the session and the timeout classifier can see it, (2) the cross-session EventType
+        /// index row, and (3) an <c>AgentEmergencyBreak</c> ops event for operator visibility
+        /// (tasks/enrollment-status-reclassification.md). Static seam with explicit dependencies so
+        /// tests can pin the artifact set without booting the Functions HTTP stack.
+        /// Best-effort — a failure here must never turn the always-200 emergency channel into a
+        /// retry loop.
+        /// </summary>
+        internal static async Task MaterializeEmergencyBreakArtifactsAsync(
+            AgentErrorReport report,
+            string tenantId,
+            ISessionRepository sessionRepo,
+            OpsEventService opsEventService,
+            ILogger logger)
+        {
+            if (report.ErrorType != AgentErrorType.SessionAgeEmergencyBreak
+                || string.IsNullOrEmpty(report.SessionId))
+            {
+                return;
+            }
+
+            try
+            {
+                var existing = await sessionRepo.GetSessionEventsAsync(tenantId, report.SessionId, maxResults: 1000);
+                // Idempotency: the agent's emergency channel can send up to a few reports per session;
+                // only ever materialize one timeline event (and one ops event).
+                var alreadyMaterialized = existing.Any(e =>
+                    string.Equals(e.EventType, Constants.EventTypes.AgentEmergencyBreak, StringComparison.OrdinalIgnoreCase));
+                if (alreadyMaterialized)
+                {
+                    return;
+                }
+
+                var evt = BuildAgentEmergencyBreakEvent(report, tenantId, existing, DateTime.UtcNow);
+                await sessionRepo.StoreEventsBatchAsync(new List<EnrollmentEvent> { evt });
+
+                // StoreEventsBatchAsync writes the Events partition only — the cross-session
+                // EventType index is normally written by EventIngestProcessor, which this
+                // backend-materialized event never passes through. Without the upsert the event
+                // exists on the session timeline but is invisible to every search-by-eventType
+                // surface (portal cross-session search, MCP search_sessions_by_event /
+                // query_raw_events) — found the hard way in the 2026-07-22 incident analysis.
+                await sessionRepo.UpsertEventTypeIndexBatchAsync(
+                    tenantId, report.SessionId, new List<EnrollmentEvent> { evt });
+
+                // Operator visibility: an emergency break means an agent silently gave up at its
+                // absolute age cap — exactly the "are we losing agents?" signal. Emitted only on
+                // first materialization so repeat reports for the same session cannot flood the
+                // ops feed. OpsEventService never throws.
+                await opsEventService.RecordAgentEmergencyBreakAsync(
+                    tenantId, report.SessionId, report.AgentVersion, evt.Message);
+            }
+            catch (Exception ex)
+            {
+                logger.LogWarning(ex,
+                    "ReportAgentError: failed to materialize agent_emergency_break event for session {SessionId}", report.SessionId);
+            }
+        }
+
         /// <summary>
         /// Builds the backend-materialized <c>agent_emergency_break</c> timeline event from the agent's
         /// best-effort emergency report. Static + pure (analog to

src/Backend/AutopilotMonitor.Functions/Services/MaintenanceService.cs

@@ -439,6 +439,13 @@ namespace AutopilotMonitor.Functions.Services
                                     var timeoutEvent = BuildSessionTimeoutEvent(session, timeoutHours, sessionEvents, now);
                                     await _sessionRepo.StoreEventsBatchAsync(new List<EnrollmentEvent> { timeoutEvent });
 
+                                    // Backend-materialized events bypass EventIngestProcessor and thus the
+                                    // cross-session EventType index — without this upsert session_timeout is
+                                    // invisible to every search-by-eventType surface (portal cross-session
+                                    // search, MCP search_sessions_by_event / query_raw_events).
+                                    await _sessionRepo.UpsertEventTypeIndexBatchAsync(
+                                        session.TenantId, session.SessionId, new List<EnrollmentEvent> { timeoutEvent });
+
                                     // StoreEventsBatchAsync only bumps the orphan side-index, not the
                                     // session's EventCount, and the terminal reconcile already ran inside
                                     // UpdateSessionStatusAsync *before* this synthetic event existed. Recount

src/Backend/AutopilotMonitor.Functions/Services/OpsEventService.cs

@@ -493,6 +493,20 @@ namespace AutopilotMonitor.Functions.Services
                 $"{sessionCount} session(s) timed out after {timeoutHours}h",
                 tenantId, "System.Maintenance", new { sessionCount, timeoutHours });
 
+        /// <summary>
+        /// An agent reported its absolute session-age emergency break (48h cap) over the
+        /// emergency channel — it cleaned itself up and exited on a session that never reached
+        /// a terminal state. This is the "are we silently losing agents?" signal; emitted by
+        /// <see cref="Functions.Ingest.ReportAgentErrorFunction"/> once per session (guarded by
+        /// the timeline-event idempotency check). Warning-tier so operators can wire a Telegram
+        /// rule; if real-world volume turns out noisy, downgrade or remove — the timeline event
+        /// on the session is the durable record.
+        /// </summary>
+        public Task RecordAgentEmergencyBreakAsync(string tenantId, string sessionId, string? agentVersion, string message)
+            => WriteAsync(OpsEventCategory.Agent, "AgentEmergencyBreak", OpsEventSeverity.Warning,
+                $"Agent emergency break on session {sessionId} (agent {agentVersion ?? "?"}): {message}",
+                tenantId, "System.EmergencyChannel", new { sessionId, agentVersion });
+
         public Task RecordExcessiveSessionEventsAsync(string tenantId, string sessionId, int eventCount, int threshold)
             => WriteAsync(OpsEventCategory.Agent, "ExcessiveSessionEvents", OpsEventSeverity.Warning,
                 $"Session {sessionId} has {eventCount} events (threshold {threshold}) — likely agent loop bug",

src/Web/autopilot-monitor-web/app/admin/components/OpsAlertRulesSection.tsx

@@ -95,7 +95,17 @@ const OPS_EVENT_TYPES: Record<string, string[]> = {
     "TenantTrialExpiring",
     "TenantTrialExpired",
   ],
-  Agent: ["BlobStorageMissing", "BlobStorageUnreachable", "NewImeVersionDetected", "ExcessiveSessionEvents"],
+  Agent: [
+    "BlobStorageMissing",
+    "BlobStorageUnreachable",
+    "NewImeVersionDetected",
+    "ExcessiveSessionEvents",
+    // 48h session-age emergency break reported over the agent's emergency channel — the
+    // "are we silently losing agents?" signal. Emitted once per session by
+    // ReportAgentErrorFunction (RecordAgentEmergencyBreakAsync). Dual-register per memory
+    // feedback_ops_event_types_dual_register.
+    "AgentEmergencyBreak",
+  ],
 };
 
 const SEVERITIES = ["Info", "Warning", "Error", "Critical"];