fix(notifications): list sort index + CAS operator Retry/Discard (P5-3, P5-4)

P5-3: add IX_Notifications_Created_NotificationId (CreatedAt DESC,
NotificationId DESC) — the Central UI Notification Outbox list page sorts on
exactly that pair, but no existing index leads unconditionally on CreatedAt,
so an unfiltered/lightly filtered page fell back to a full clustered-index
scan + sort on every COUNT + page request. Mirrors the Medium wave's
IX_SiteCalls_Created_TrackedOperationId migration + SQL twin pattern
faithfully, including the handwritten docs/plans/sql script.

P5-4: operator Retry/Discard was read-check-write keyed on NotificationId
only — a detached PipeTo task interleaving with the dispatcher's own write
(or a second concurrent operator action) could land between the read and
the write and get silently clobbered by the unconditional UpdateAsync,
stamping a Delivered row Discarded with stale fields. Adds
INotificationOutboxRepository.TryUpdateFromParkedAsync, a single-statement
compare-and-swap (WHERE Status == Parked) mirroring
SecuredWriteRepository.TryMarkApprovedAsync's CAS shape. NotificationOutboxActor's
RetryAsync/DiscardAsync now use the CAS write and, on a lost race, re-read
to report "notification already progressed to <Status>" instead of silently
overwriting whatever is there now.

Proof of failure: reverting only the actor's CAS usage while keeping the
repository addition reproduces the silent-overwrite bug — 7 tests fail,
including two new deterministic (non-sleep) race-proof tests that mock the
exact GetByIdAsync -> lost-CAS -> re-read sequence the finding describes.
P5-3 is perf-only; verified via migration/SQL-twin parity instead (no
runtime proof-of-failure is meaningful for an index).

docs/plans/sql/AddNotificationsListSortIndex.sql

@@ -0,0 +1,51 @@
+-- arch-review 2026-08-19 P5-3 (register row-35 class): the Notifications list
+-- page's sort key gets its own index.
+--
+-- NotificationOutboxRepository.QueryAsync orders (CreatedAt DESC, NotificationId
+-- DESC) (NotificationId breaks CreatedAt ties for deterministic OFFSET paging).
+-- None of the existing Notifications indexes lead unconditionally on CreatedAt:
+--   - IX_Notifications_Status_NextAttemptAt requires an equality on Status first.
+--   - IX_Notifications_SourceSiteId_CreatedAt requires an equality on SourceSiteId first.
+--   - IX_Notifications_Delivered is filtered to Status = 'Delivered' only.
+-- So an unfiltered (or lightly filtered) page fell back to a full clustered-index
+-- scan + sort. IX_Notifications_Created_NotificationId (CreatedAt DESC,
+-- NotificationId DESC) makes a forward scan of the index deliver rows in the
+-- requested order directly, so a page reads a Top-N off the index instead of
+-- sorting the whole table on every COUNT + page request.
+--
+-- Notifications is small enough today that a plain (offline) index build is
+-- fine. On an Enterprise edition instance where the table has grown large and
+-- the Central UI must stay responsive, add WITH (ONLINE = ON) to the CREATE
+-- INDEX below.
+--
+-- Run with QUOTED_IDENTIFIER ON. Several tables in this database carry filtered
+-- indexes and persisted computed columns, and SQL Server refuses DDL/DML against
+-- them with error 1934 when the session has QUOTED_IDENTIFIER OFF -- which is
+-- exactly what `sqlcmd` gives you by default (pass -I, or keep this header). The
+-- SETs sit in their own batch so they are in force when the statements after the
+-- GO parse.
+SET QUOTED_IDENTIFIER ON;
+SET ANSI_NULLS ON;
+GO
+
+BEGIN TRANSACTION;
+IF NOT EXISTS (
+    SELECT * FROM [__EFMigrationsHistory]
+    WHERE [MigrationId] = N'20260819201210_AddNotificationsListSortIndex'
+)
+BEGIN
+    EXEC(N'CREATE INDEX [IX_Notifications_Created_NotificationId]
+        ON [Notifications] ([CreatedAt] DESC, [NotificationId] DESC)');
+END;
+
+IF NOT EXISTS (
+    SELECT * FROM [__EFMigrationsHistory]
+    WHERE [MigrationId] = N'20260819201210_AddNotificationsListSortIndex'
+)
+BEGIN
+    INSERT INTO [__EFMigrationsHistory] ([MigrationId], [ProductVersion])
+    VALUES (N'20260819201210_AddNotificationsListSortIndex', N'10.0.7');
+END;
+
+COMMIT;
+GO

src/ZB.MOM.WW.ScadaBridge.Commons/Interfaces/Repositories/INotificationOutboxRepository.cs

@@ -1,4 +1,5 @@
 using ZB.MOM.WW.ScadaBridge.Commons.Entities.Notifications;
+using ZB.MOM.WW.ScadaBridge.Commons.Types.Enums;
 using ZB.MOM.WW.ScadaBridge.Commons.Types.Notifications;
 
 namespace ZB.MOM.WW.ScadaBridge.Commons.Interfaces.Repositories;
@@ -83,6 +84,43 @@ public interface INotificationOutboxRepository
     /// </returns>
     Task<bool> UpdateAsync(Notification n, CancellationToken cancellationToken = default);
 
+    /// <summary>
+    /// Compare-and-swap variant of <see cref="UpdateAsync"/> for the operator
+    /// Retry/Discard one-shots: persists <paramref name="n"/>'s delivery-state
+    /// columns (same narrow column set as <see cref="UpdateAsync"/>) ONLY if the
+    /// row's current <c>Status</c> in the database is still
+    /// <see cref="NotificationStatus.Parked"/>. Commits internally — this call
+    /// is its own transaction.
+    /// </summary>
+    /// <remarks>
+    /// <b>Why this exists (arch-review 2026-08-19 P5-4).</b> The operator retry/
+    /// discard handlers are read-check-write across two round trips:
+    /// <see cref="GetByIdAsync"/> reads the row, the handler checks its
+    /// in-memory <c>Status</c>, then a write persists the transition. A detached
+    /// <c>PipeTo</c> task can interleave with the dispatcher's own write (or a
+    /// second concurrent operator action) between those two round trips, so the
+    /// in-memory check is a fast-path rejection, not a correctness guard — a
+    /// blind <see cref="UpdateAsync"/> in that window can stamp a row that has
+    /// already moved on (e.g. the dispatcher delivered it) with the operator's
+    /// stale fields, silently overwriting a <c>Delivered</c> row as
+    /// <c>Discarded</c>. The <c>WHERE Status = Parked</c> predicate below makes
+    /// the <c>Parked-&gt;(Pending|Discarded)</c> transition atomic at the row
+    /// level, mirroring <c>SecuredWriteRepository.TryMarkApprovedAsync</c>'s
+    /// single-statement compare-and-swap. A row that has already progressed
+    /// (dispatcher won the race, or a second operator action won it) and a row
+    /// that has vanished (retention purge) both report zero rows affected;
+    /// callers cannot tell those apart from the return value alone and must
+    /// re-read to report which honestly.
+    /// </remarks>
+    /// <param name="n">The notification whose delivery state should be persisted, if still Parked.</param>
+    /// <param name="cancellationToken">Cancellation token.</param>
+    /// <returns>
+    /// <see langword="true"/> when a row with that <c>NotificationId</c> was
+    /// found with <c>Status == Parked</c> and its delivery state was persisted;
+    /// <see langword="false"/> otherwise (not found, or no longer Parked).
+    /// </returns>
+    Task<bool> TryUpdateFromParkedAsync(Notification n, CancellationToken cancellationToken = default);
+
     /// <summary>
     /// Returns a page of notifications matching <paramref name="filter"/>, ordered by
     /// <c>CreatedAt</c> descending, together with the total matching count.

src/ZB.MOM.WW.ScadaBridge.ConfigurationDatabase/Configurations/NotificationOutboxConfiguration.cs

@@ -83,5 +83,23 @@ public class NotificationOutboxConfiguration : IEntityTypeConfiguration<Notifica
         builder.HasIndex(n => n.DeliveredAt)
             .HasDatabaseName("IX_Notifications_Delivered")
             .HasFilter("[Status] = 'Delivered'");
+
+        // Created_NotificationId backs the Central UI Notification Outbox list
+        // page's sort key — NotificationOutboxRepository.QueryAsync orders
+        // (CreatedAt DESC, NotificationId DESC) (NotificationId breaks CreatedAt
+        // ties for deterministic OFFSET paging). None of the indexes above lead
+        // unconditionally on CreatedAt: Status_NextAttemptAt requires an equality
+        // on Status first, SourceSiteId_CreatedAt requires an equality on
+        // SourceSiteId first, and IX_Notifications_Delivered is filtered to
+        // Status='Delivered' only. So an unfiltered (or lightly filtered) page
+        // fell back to a full clustered-index scan + sort on every COUNT + page
+        // request (arch-review 2026-08-19 P5-3, register row-35 class). With both
+        // key columns descending, a forward scan of this index already delivers
+        // rows in the requested order. Exact mirror of
+        // SiteCallEntityTypeConfiguration's IX_SiteCalls_Created_TrackedOperationId
+        // (2026-08-19, arch-review P5-2).
+        builder.HasIndex(n => new { n.CreatedAt, n.NotificationId })
+            .IsDescending(true, true)
+            .HasDatabaseName("IX_Notifications_Created_NotificationId");
     }
 }

src/ZB.MOM.WW.ScadaBridge.ConfigurationDatabase/Migrations/20260819201210_AddNotificationsListSortIndex.cs

@@ -0,0 +1,46 @@
+using Microsoft.EntityFrameworkCore.Migrations;
+
+#nullable disable
+
+namespace ZB.MOM.WW.ScadaBridge.ConfigurationDatabase.Migrations
+{
+    /// <summary>
+    /// Adds <c>IX_Notifications_Created_NotificationId (CreatedAt DESC,
+    /// NotificationId DESC)</c> — arch-review 2026-08-19 P5-3, register
+    /// row-35 class.
+    /// </summary>
+    /// <remarks>
+    /// <see cref="ZB.MOM.WW.ScadaBridge.ConfigurationDatabase.Repositories.NotificationOutboxRepository.QueryAsync"/>
+    /// orders <c>(CreatedAt DESC, NotificationId DESC)</c> for the Central UI
+    /// Notification Outbox list page. None of the existing Notifications indexes
+    /// lead unconditionally on <c>CreatedAt</c> — <c>IX_Notifications_Status_NextAttemptAt</c>
+    /// requires an equality on <c>Status</c> first, <c>IX_Notifications_SourceSiteId_CreatedAt</c>
+    /// requires an equality on <c>SourceSiteId</c> first, and
+    /// <c>IX_Notifications_Delivered</c> is filtered to <c>Status = 'Delivered'</c>
+    /// only — so an unfiltered (or lightly filtered) list page fell back to a full
+    /// clustered-index scan + sort on every COUNT + page request. With both key
+    /// columns descending, a forward scan of this index already delivers rows in
+    /// the requested order. See <c>docs/plans/sql/AddNotificationsListSortIndex.sql</c>
+    /// for the handwritten production-parity script.
+    /// </remarks>
+    public partial class AddNotificationsListSortIndex : Migration
+    {
+        /// <inheritdoc />
+        protected override void Up(MigrationBuilder migrationBuilder)
+        {
+            migrationBuilder.CreateIndex(
+                name: "IX_Notifications_Created_NotificationId",
+                table: "Notifications",
+                columns: new[] { "CreatedAt", "NotificationId" },
+                descending: new bool[0]);
+        }
+
+        /// <inheritdoc />
+        protected override void Down(MigrationBuilder migrationBuilder)
+        {
+            migrationBuilder.DropIndex(
+                name: "IX_Notifications_Created_NotificationId",
+                table: "Notifications");
+        }
+    }
+}

src/ZB.MOM.WW.ScadaBridge.ConfigurationDatabase/Migrations/ScadaBridgeDbContextModelSnapshot.cs

@@ -913,6 +913,10 @@ namespace ZB.MOM.WW.ScadaBridge.ConfigurationDatabase.Migrations
                         .HasDatabaseName("IX_Notifications_Delivered")
                         .HasFilter("[Status] = 'Delivered'");
 
+                    b.HasIndex("CreatedAt", "NotificationId")
+                        .IsDescending()
+                        .HasDatabaseName("IX_Notifications_Created_NotificationId");
+
                     b.HasIndex("SourceSiteId", "CreatedAt");
 
                     b.HasIndex("Status", "NextAttemptAt");

src/ZB.MOM.WW.ScadaBridge.ConfigurationDatabase/Repositories/NotificationOutboxRepository.cs

@@ -205,6 +205,48 @@ VALUES
         return rowsAffected > 0;
     }
 
+    /// <inheritdoc />
+    public async Task<bool> TryUpdateFromParkedAsync(Notification n, CancellationToken cancellationToken = default)
+    {
+        ArgumentNullException.ThrowIfNull(n);
+
+        // Same seven mutable delivery-state columns as UpdateAsync, but the WHERE
+        // clause below ALSO requires the row's current Status to still be Parked
+        // -- a single-statement compare-and-swap (arch-review 2026-08-19 P5-4).
+        // Two round trips (GetByIdAsync then this write) with a status check in
+        // between is not a correctness guard by itself: a detached PipeTo task
+        // can race the dispatcher's own write (or a second concurrent operator
+        // action) in that window. Mirrors SecuredWriteRepository.TryMarkApprovedAsync's
+        // conditional ExecuteUpdate/ExecuteSql pattern, expressed via LINQ
+        // ExecuteUpdateAsync (like UpdateAsync above) so this stays portable
+        // across the SQL Server production provider and the SQLite provider the
+        // repository tests run against.
+        var status = n.Status;
+        var retryCount = n.RetryCount;
+        var lastError = n.LastError;
+        var resolvedTargets = n.ResolvedTargets;
+        var lastAttemptAt = n.LastAttemptAt;
+        var nextAttemptAt = n.NextAttemptAt;
+        var deliveredAt = n.DeliveredAt;
+
+        var notificationId = n.NotificationId;
+
+        var rowsAffected = await _context.Notifications
+            .Where(row => row.NotificationId == notificationId && row.Status == NotificationStatus.Parked)
+            .ExecuteUpdateAsync(
+                setters => setters
+                    .SetProperty(row => row.Status, status)
+                    .SetProperty(row => row.RetryCount, retryCount)
+                    .SetProperty(row => row.LastError, lastError)
+                    .SetProperty(row => row.ResolvedTargets, resolvedTargets)
+                    .SetProperty(row => row.LastAttemptAt, lastAttemptAt)
+                    .SetProperty(row => row.NextAttemptAt, nextAttemptAt)
+                    .SetProperty(row => row.DeliveredAt, deliveredAt),
+                cancellationToken);
+
+        return rowsAffected > 0;
+    }
+
     /// <inheritdoc />
     public async Task<Notification?> GetByIdAsync(string notificationId, CancellationToken cancellationToken = default)
     {

src/ZB.MOM.WW.ScadaBridge.NotificationOutbox/NotificationOutboxActor.cs

@@ -1109,14 +1109,25 @@ public class NotificationOutboxActor : ReceiveActor, IWithTimers
         notification.NextAttemptAt = null;
         notification.LastError = null;
 
-        // Zero rows updated means the row was purged between the read above and
-        // this write. Answer the operator honestly instead of reporting a
-        // re-queue that never happened — and emit no un-park audit row, because
-        // there is nothing to attribute it to.
-        if (!await repository.UpdateAsync(notification))
+        // Compare-and-swap on Status == Parked (arch-review 2026-08-19 P5-4). The
+        // read above and this write are two separate round trips; a detached
+        // PipeTo task can race the dispatcher's own write (or a second
+        // concurrent operator action) in between, so the in-memory check above
+        // is a fast-path rejection, not the correctness guard — a blind
+        // UpdateAsync here could stamp a row that has already moved on (e.g. the
+        // dispatcher delivered it) back to Pending with stale fields. A lost CAS
+        // means either the row vanished (purge) or it progressed past Parked
+        // under us; re-read to report which honestly instead of silently
+        // overwriting whatever is there now, and emit no un-park audit row in
+        // either case, because there is nothing to attribute it to.
+        if (!await repository.TryUpdateFromParkedAsync(notification))
         {
+            var current = await repository.GetByIdAsync(request.NotificationId);
             return new RetryNotificationResponse(
-                request.CorrelationId, Success: false, ErrorMessage: "notification not found");
+                request.CorrelationId, Success: false,
+                ErrorMessage: current is null
+                    ? "notification not found"
+                    : $"notification already progressed to {current.Status}; retry not applied");
         }
 
         // Operator re-queued a parked notification. Emit a Submitted NotifyDeliver
@@ -1190,13 +1201,22 @@ public class NotificationOutboxActor : ReceiveActor, IWithTimers
 
         notification.Status = NotificationStatus.Discarded;
 
-        // Same vanished-row honesty as the retry path: a purge between the read
-        // and the write leaves nothing to discard, so report not-found rather
-        // than success (and skip the terminal audit row below).
-        if (!await repository.UpdateAsync(notification))
+        // Same compare-and-swap on Status == Parked as the retry path (arch-review
+        // 2026-08-19 P5-4): a purge between the read and the write leaves nothing
+        // to discard, and a dispatcher/other-operator write landing in that same
+        // window (e.g. a delivery succeeding) must not be silently clobbered by a
+        // blind write of this stale in-memory copy. Re-read on a lost CAS to
+        // report which case it was, and skip the terminal audit row below either
+        // way — there is nothing to attribute a discard of a row that was never
+        // actually discarded.
+        if (!await repository.TryUpdateFromParkedAsync(notification))
         {
+            var current = await repository.GetByIdAsync(request.NotificationId);
             return new DiscardNotificationResponse(
-                request.CorrelationId, Success: false, ErrorMessage: "notification not found");
+                request.CorrelationId, Success: false,
+                ErrorMessage: current is null
+                    ? "notification not found"
+                    : $"notification already progressed to {current.Status}; discard not applied");
         }
 
         // A manual discard is the OTHER code path that produces

tests/ZB.MOM.WW.ScadaBridge.ConfigurationDatabase.Tests/RepositoryCoverageTests.cs

@@ -573,6 +573,154 @@ public class NotificationOutboxRepositoryTests : IDisposable
         Assert.False(await _repository.UpdateAsync(loaded));
     }
 
+    [Fact]
+    public async Task TryUpdateFromParkedAsync_StatusIsParked_UpdatesAndReturnsTrue()
+    {
+        var id = Guid.NewGuid().ToString();
+        _context.Notifications.Add(MakeNotification(id, NotificationStatus.Parked));
+        await _context.SaveChangesAsync();
+        _context.ChangeTracker.Clear();
+
+        var loaded = await _repository.GetByIdAsync(id);
+        loaded!.Status = NotificationStatus.Pending;
+        loaded.RetryCount = 0;
+        loaded.NextAttemptAt = null;
+        loaded.LastError = null;
+        Assert.True(await _repository.TryUpdateFromParkedAsync(loaded));
+
+        _context.ChangeTracker.Clear();
+        var reloaded = await _context.Notifications.FindAsync(id);
+        Assert.Equal(NotificationStatus.Pending, reloaded!.Status);
+    }
+
+    /// <summary>
+    /// Direct CAS-miss: the row is currently Delivered (never Parked in this
+    /// scenario), so the conditional WHERE Status == Parked matches nothing and
+    /// the row is left completely untouched — arch-review 2026-08-19 P5-4.
+    /// </summary>
+    [Fact]
+    public async Task TryUpdateFromParkedAsync_StatusIsNotParked_ReturnsFalse_AndLeavesRowUnchanged()
+    {
+        var id = Guid.NewGuid().ToString();
+        var deliveredAt = new DateTimeOffset(2026, 5, 19, 9, 0, 0, TimeSpan.Zero);
+        _context.Notifications.Add(MakeNotification(
+            id, NotificationStatus.Delivered, deliveredAt: deliveredAt));
+        await _context.SaveChangesAsync();
+        _context.ChangeTracker.Clear();
+
+        var staleDiscard = MakeNotification(id, NotificationStatus.Discarded);
+        Assert.False(await _repository.TryUpdateFromParkedAsync(staleDiscard));
+
+        _context.ChangeTracker.Clear();
+        var reloaded = await _context.Notifications.FindAsync(id);
+        Assert.Equal(NotificationStatus.Delivered, reloaded!.Status);
+        Assert.Equal(deliveredAt, reloaded.DeliveredAt);
+    }
+
+    [Fact]
+    public async Task TryUpdateFromParkedAsync_RowPurgedBeforeWrite_ReturnsFalse()
+    {
+        var id = Guid.NewGuid().ToString();
+        _context.Notifications.Add(MakeNotification(id, NotificationStatus.Parked));
+        await _context.SaveChangesAsync();
+        _context.ChangeTracker.Clear();
+
+        var loaded = await _repository.GetByIdAsync(id);
+        Assert.NotNull(loaded);
+
+        await _context.Notifications.Where(n => n.NotificationId == id).ExecuteDeleteAsync();
+        _context.ChangeTracker.Clear();
+
+        loaded!.Status = NotificationStatus.Discarded;
+        Assert.False(await _repository.TryUpdateFromParkedAsync(loaded));
+    }
+
+    /// <summary>
+    /// Deterministic (non-sleep) proof of the arch-review 2026-08-19 P5-4 race:
+    /// the operator reads a Parked row, but before its write lands the
+    /// dispatcher (or a concurrent second operator action) has already moved
+    /// the row to Delivered. <see cref="TryUpdateFromParkedAsync"/>'s CAS must
+    /// refuse the stale write and leave the Delivered row exactly as it is.
+    /// The companion fact below proves the general-purpose
+    /// <see cref="NotificationOutboxRepository.UpdateAsync"/> — the method the
+    /// operator path used PRE-FIX — has no such guard, which is the mechanism
+    /// the finding describes.
+    /// </summary>
+    [Fact]
+    public async Task TryUpdateFromParkedAsync_ConcurrentlyDelivered_ReturnsFalse_AndDoesNotOverwrite()
+    {
+        var id = Guid.NewGuid().ToString();
+        _context.Notifications.Add(MakeNotification(id, NotificationStatus.Parked));
+        await _context.SaveChangesAsync();
+        _context.ChangeTracker.Clear();
+
+        // The operator's read (this snapshot is carried forward stale below).
+        var operatorRead = await _repository.GetByIdAsync(id);
+        Assert.NotNull(operatorRead);
+
+        // The dispatcher's own write lands first, in the window between the
+        // operator's read and write.
+        var deliveredAt = new DateTimeOffset(2026, 5, 19, 9, 0, 0, TimeSpan.Zero);
+        var dispatcherRead = await _repository.GetByIdAsync(id);
+        dispatcherRead!.Status = NotificationStatus.Delivered;
+        dispatcherRead.DeliveredAt = deliveredAt;
+        Assert.True(await _repository.UpdateAsync(dispatcherRead));
+        _context.ChangeTracker.Clear();
+
+        // The operator's stale in-memory copy still says Parked; it attempts a Discard.
+        operatorRead!.Status = NotificationStatus.Discarded;
+        Assert.False(await _repository.TryUpdateFromParkedAsync(operatorRead));
+
+        _context.ChangeTracker.Clear();
+        var reloaded = await _context.Notifications.FindAsync(id);
+        Assert.Equal(NotificationStatus.Delivered, reloaded!.Status);
+        Assert.Equal(deliveredAt, reloaded.DeliveredAt);
+    }
+
+    /// <summary>
+    /// Companion to <see cref="TryUpdateFromParkedAsync_ConcurrentlyDelivered_ReturnsFalse_AndDoesNotOverwrite"/>:
+    /// runs the IDENTICAL interleaving through the unconditional
+    /// <see cref="NotificationOutboxRepository.UpdateAsync"/> instead — the
+    /// method the operator Retry/Discard path called pre-fix — and shows it
+    /// silently clobbers the Delivered row. This documents the exact defect
+    /// arch-review 2026-08-19 P5-4 fixed by routing the operator path through
+    /// <see cref="TryUpdateFromParkedAsync"/>: UpdateAsync itself is unchanged
+    /// (the dispatcher still needs an unconditional write), so this scenario
+    /// remains "true" by design — it is the operator actor's call-site choice
+    /// that was the bug, proven fixed by the actor-level tests in
+    /// NotificationOutbox.Tests (Discard_ConcurrentlyDelivered_...).
+    /// </summary>
+    [Fact]
+    public async Task UpdateAsync_Unconditional_WouldOverwriteConcurrentlyDeliveredRow_DemonstratingP5_4Mechanism()
+    {
+        var id = Guid.NewGuid().ToString();
+        _context.Notifications.Add(MakeNotification(id, NotificationStatus.Parked));
+        await _context.SaveChangesAsync();
+        _context.ChangeTracker.Clear();
+
+        var operatorRead = await _repository.GetByIdAsync(id);
+        Assert.NotNull(operatorRead);
+
+        var deliveredAt = new DateTimeOffset(2026, 5, 19, 9, 0, 0, TimeSpan.Zero);
+        var dispatcherRead = await _repository.GetByIdAsync(id);
+        dispatcherRead!.Status = NotificationStatus.Delivered;
+        dispatcherRead.DeliveredAt = deliveredAt;
+        Assert.True(await _repository.UpdateAsync(dispatcherRead));
+        _context.ChangeTracker.Clear();
+
+        operatorRead!.Status = NotificationStatus.Discarded;
+        // The unconditional UPDATE has no status guard, so it "succeeds" here —
+        // this is the bug. It clobbers the Delivered row to Discarded and drops
+        // DeliveredAt, which is exactly the silent-overwrite hazard P5-4 closes
+        // by moving the operator path to TryUpdateFromParkedAsync.
+        Assert.True(await _repository.UpdateAsync(operatorRead));
+
+        _context.ChangeTracker.Clear();
+        var reloaded = await _context.Notifications.FindAsync(id);
+        Assert.Equal(NotificationStatus.Discarded, reloaded!.Status);
+        Assert.Null(reloaded.DeliveredAt);
+    }
+
     [Fact]
     public async Task QueryAsync_AppliesFilters_OrdersByCreatedAtDescending_AndPaginates()
     {

tests/ZB.MOM.WW.ScadaBridge.NotificationOutbox.Tests/NotificationOutboxActorQueryTests.cs

@@ -258,7 +258,7 @@ public class NotificationOutboxActorQueryTests : TestKit
         Assert.True(response.Success);
         Assert.Null(response.ErrorMessage);
 
-        _repository.Received(1).UpdateAsync(
+        _repository.Received(1).TryUpdateFromParkedAsync(
             Arg.Is<Notification>(n =>
                 n.NotificationId == row.NotificationId &&
                 n.Status == NotificationStatus.Pending &&
@@ -280,7 +280,7 @@ public class NotificationOutboxActorQueryTests : TestKit
         var response = ExpectMsg<RetryNotificationResponse>();
         Assert.False(response.Success);
         Assert.NotNull(response.ErrorMessage);
-        _repository.DidNotReceive().UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
+        _repository.DidNotReceive().TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
     }
 
     [Fact]
@@ -300,17 +300,19 @@ public class NotificationOutboxActorQueryTests : TestKit
 
     /// <summary>
     /// The row is read successfully but the retention purge deletes it before the
-    /// write lands, so the targeted UPDATE matches nothing. The operator must be
-    /// told the notification is gone — reporting a re-queue that never happened
-    /// is what the ExecuteUpdate rewrite silently introduced (the predecessor
-    /// threw DbUpdateConcurrencyException here).
+    /// write lands, so the CAS write matches nothing. The re-read that follows a
+    /// lost CAS (arch-review 2026-08-19 P5-4) finds nothing either, and the
+    /// operator must be told the notification is gone — reporting a re-queue
+    /// that never happened is what a blind ExecuteUpdate would silently do (the
+    /// pre-ExecuteUpdate predecessor threw DbUpdateConcurrencyException here).
     /// </summary>
     [Fact]
     public void Retry_RowPurgedBeforeWrite_RepliesNotFound()
     {
         var row = MakeNotification(status: NotificationStatus.Parked, retryCount: 10, lastError: "gave up");
-        _repository.GetByIdAsync(row.NotificationId, Arg.Any<CancellationToken>()).Returns(row);
-        _repository.UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
+        _repository.GetByIdAsync(row.NotificationId, Arg.Any<CancellationToken>())
+            .Returns(row, (Notification?)null);
+        _repository.TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
         var actor = CreateActor();
 
         actor.Tell(new RetryNotificationRequest("corr-vanished-retry", row.NotificationId), TestActor);
@@ -327,8 +329,9 @@ public class NotificationOutboxActorQueryTests : TestKit
     public void Discard_RowPurgedBeforeWrite_RepliesNotFound()
     {
         var row = MakeNotification(status: NotificationStatus.Parked);
-        _repository.GetByIdAsync(row.NotificationId, Arg.Any<CancellationToken>()).Returns(row);
-        _repository.UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
+        _repository.GetByIdAsync(row.NotificationId, Arg.Any<CancellationToken>())
+            .Returns(row, (Notification?)null);
+        _repository.TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
         var actor = CreateActor();
 
         actor.Tell(new DiscardNotificationRequest("corr-vanished-discard", row.NotificationId), TestActor);
@@ -340,6 +343,75 @@ public class NotificationOutboxActorQueryTests : TestKit
         Assert.Contains("not found", response.ErrorMessage);
     }
 
+    /// <summary>
+    /// PROOF OF FAILURE for arch-review 2026-08-19 P5-4. Models the race the
+    /// finding describes: the operator's read sees Parked, but by the time the
+    /// write lands the dispatcher has already delivered the row (a detached
+    /// PipeTo task interleaving with the dispatcher's own write). The mechanism
+    /// is proven deterministically via a mocked call sequence, not a wall-clock
+    /// sleep (per the "no new sleep-then-assert" constraint) — the first
+    /// <see cref="INotificationOutboxRepository.GetByIdAsync"/> call (the
+    /// operator's read) returns the Parked row, the CAS write reports a lost
+    /// race, and the SECOND <c>GetByIdAsync</c> call (the fix's re-read used to
+    /// build an honest reply) returns the row now Delivered.
+    /// <para>
+    /// Against the pre-fix actor — which calls the unconditional
+    /// <c>UpdateAsync</c> exactly once and never re-reads — this test fails:
+    /// the single-read code path only ever consumes the first queued value
+    /// (Parked), the in-memory check passes, <c>UpdateAsync</c> is stubbed
+    /// <c>Healthy()</c> (returns true), and the response comes back
+    /// <c>Success = true</c> — i.e. the Delivered row gets silently clobbered
+    /// to Discarded. Post-fix, the CAS is lost, the actor re-reads and reports
+    /// "already progressed to Delivered" instead of overwriting it.
+    /// </para>
+    /// </summary>
+    [Fact]
+    public void Discard_ConcurrentlyDelivered_RepliesAlreadyProgressed_DoesNotOverwrite()
+    {
+        var parkedRow = MakeNotification(status: NotificationStatus.Parked, retryCount: 5, lastError: "gave up");
+        var deliveredRow = MakeNotification(status: NotificationStatus.Delivered);
+        _repository.GetByIdAsync(parkedRow.NotificationId, Arg.Any<CancellationToken>())
+            .Returns(parkedRow, deliveredRow);
+        _repository.TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
+        var actor = CreateActor();
+
+        actor.Tell(new DiscardNotificationRequest("corr-race-discard", parkedRow.NotificationId), TestActor);
+
+        var response = ExpectMsg<DiscardNotificationResponse>();
+        Assert.Equal("corr-race-discard", response.CorrelationId);
+        Assert.False(response.Success);
+        Assert.NotNull(response.ErrorMessage);
+        Assert.Contains("already progressed", response.ErrorMessage);
+        Assert.Contains("Delivered", response.ErrorMessage);
+
+        // The critical invariant: the old unconditional UpdateAsync must NEVER
+        // be reached by the operator path — that call is what silently
+        // clobbered the Delivered row pre-fix.
+        _repository.DidNotReceive().UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
+    }
+
+    /// <summary>Retry half of <see cref="Discard_ConcurrentlyDelivered_RepliesAlreadyProgressed_DoesNotOverwrite"/>.</summary>
+    [Fact]
+    public void Retry_ConcurrentlyDelivered_RepliesAlreadyProgressed_DoesNotOverwrite()
+    {
+        var parkedRow = MakeNotification(status: NotificationStatus.Parked, retryCount: 5, lastError: "gave up");
+        var deliveredRow = MakeNotification(status: NotificationStatus.Delivered);
+        _repository.GetByIdAsync(parkedRow.NotificationId, Arg.Any<CancellationToken>())
+            .Returns(parkedRow, deliveredRow);
+        _repository.TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>()).Returns(false);
+        var actor = CreateActor();
+
+        actor.Tell(new RetryNotificationRequest("corr-race-retry", parkedRow.NotificationId), TestActor);
+
+        var response = ExpectMsg<RetryNotificationResponse>();
+        Assert.Equal("corr-race-retry", response.CorrelationId);
+        Assert.False(response.Success);
+        Assert.NotNull(response.ErrorMessage);
+        Assert.Contains("already progressed", response.ErrorMessage);
+        Assert.Contains("Delivered", response.ErrorMessage);
+        _repository.DidNotReceive().UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
+    }
+
     [Fact]
     public void Discard_ParkedNotification_MarksDiscarded_AndSucceeds()
     {
@@ -353,7 +425,7 @@ public class NotificationOutboxActorQueryTests : TestKit
         Assert.Equal("corr-8", response.CorrelationId);
         Assert.True(response.Success);
 
-        _repository.Received(1).UpdateAsync(
+        _repository.Received(1).TryUpdateFromParkedAsync(
             Arg.Is<Notification>(n => n.Status == NotificationStatus.Discarded),
             Arg.Any<CancellationToken>());
     }
@@ -370,7 +442,7 @@ public class NotificationOutboxActorQueryTests : TestKit
         var response = ExpectMsg<DiscardNotificationResponse>();
         Assert.False(response.Success);
         Assert.NotNull(response.ErrorMessage);
-        _repository.DidNotReceive().UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
+        _repository.DidNotReceive().TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>());
     }
 
     [Fact]

tests/ZB.MOM.WW.ScadaBridge.NotificationOutbox.Tests/NotificationOutboxActorRetryEmissionTests.cs

@@ -147,7 +147,7 @@ public class NotificationOutboxActorRetryEmissionTests : TestKit
         ExpectMsg<RetryNotificationResponse>(r => r.Success);
         AwaitAssert(() =>
         {
-            _outboxRepository.Received(1).UpdateAsync(
+            _outboxRepository.Received(1).TryUpdateFromParkedAsync(
                 Arg.Is<Notification>(n => n.Status == NotificationStatus.Pending),
                 Arg.Any<CancellationToken>());
         });

tests/ZB.MOM.WW.ScadaBridge.NotificationOutbox.Tests/OutboxRepositorySubstitute.cs

@@ -8,14 +8,16 @@ namespace ZB.MOM.WW.ScadaBridge.NotificationOutbox.Tests;
 /// Factory for the outbox repository substitute used across these tests.
 /// </summary>
 /// <remarks>
-/// <see cref="INotificationOutboxRepository.UpdateAsync"/> returns whether the
-/// targeted UPDATE actually matched a row — <c>false</c> is the "this
-/// notification no longer exists" signal the operator retry/discard handlers
-/// turn into a not-found reply. NSubstitute's default for <c>Task&lt;bool&gt;</c>
-/// is <c>false</c>, i.e. the VANISHED-row answer, so an unconfigured substitute
-/// would silently put every test on the failure path. Tests that want a healthy
-/// store take one from here; the vanished-row tests configure
-/// <c>Returns(false)</c> for themselves.
+/// <see cref="INotificationOutboxRepository.UpdateAsync"/> and
+/// <see cref="INotificationOutboxRepository.TryUpdateFromParkedAsync"/> both
+/// return whether the targeted UPDATE actually matched a row — <c>false</c> is
+/// the "this notification no longer exists (or, for the CAS variant, is no
+/// longer Parked)" signal the operator retry/discard handlers turn into a
+/// not-found / already-progressed reply. NSubstitute's default for
+/// <c>Task&lt;bool&gt;</c> is <c>false</c>, i.e. the failure answer, so an
+/// unconfigured substitute would silently put every test on the failure path.
+/// Tests that want a healthy store take one from here; the vanished-row and
+/// lost-CAS tests configure <c>Returns(false)</c> for themselves.
 /// </remarks>
 internal static class OutboxRepositorySubstitute
 {
@@ -26,6 +28,9 @@ internal static class OutboxRepositorySubstitute
         repository
             .UpdateAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>())
             .Returns(true);
+        repository
+            .TryUpdateFromParkedAsync(Arg.Any<Notification>(), Arg.Any<CancellationToken>())
+            .Returns(true);
         return repository;
     }
 }