diff --git a/tests/ZB.MOM.WW.ScadaBridge.DataConnectionLayer.Tests/Actors/HighReviewRegressionTests.cs b/tests/ZB.MOM.WW.ScadaBridge.DataConnectionLayer.Tests/Actors/HighReviewRegressionTests.cs
index 761c566a..4ff588db 100644
--- a/tests/ZB.MOM.WW.ScadaBridge.DataConnectionLayer.Tests/Actors/HighReviewRegressionTests.cs
+++ b/tests/ZB.MOM.WW.ScadaBridge.DataConnectionLayer.Tests/Actors/HighReviewRegressionTests.cs
@@ -25,7 +25,6 @@ public class HighReviewRegressionTests : TestKit
     private IActorRef Actor(IDataConnection adapter) => Sys.ActorOf(Props.Create(() => new DataConnectionActor(
         "conn", adapter, new DataConnectionOptions { ReconnectInterval = TimeSpan.FromMilliseconds(50), TagResolutionRetryInterval = TimeSpan.FromHours(1), SeedReadMaxAttempts = 1 },
         _health, Substitute.For<IDataConnectionFactory>(), "OpcUa")));
-    private void Connected() => AwaitCondition(() => _health.ReceivedCalls().Any(c => c.GetMethodInfo().Name == "UpdateConnectionHealth" && c.GetArguments()[1] is ConnectionHealth.Connected));
     private static SubscribeTagsRequest Subscribe(string correlation = "s") => new(correlation, "instance", "conn", ["tag"], DateTimeOffset.UtcNow);
 
     private DataConnectionHealthReport Health(IActorRef actor)
@@ -163,6 +162,17 @@ public class HighReviewRegressionTests : TestKit
 
     /// <summary>Verifies that writes while disconnected are rejected and never replayed after reconnect.</summary>
     /// <param name="reconnect">Whether the connection is re-established after the rejected write.</param>
+    /// <remarks>
+    /// ScadaBridge#101: the writes used to be sent straight after <c>ActorOf</c> and their
+    /// replies awaited for a fixed 1 s, so in the <c>reconnect: false</c> case that window
+    /// also had to cover the actor's start and its first dispatch — a starved thread pool
+    /// in a full solution run could miss it. The test now waits on a barrier instead: the
+    /// connect attempt the writes race against has been ISSUED (the adapter's
+    /// <c>ConnectAsync</c> was entered) and is held open by <c>gate</c>, so the actor is
+    /// provably mid-attempt and cannot connect until the test releases it. The reply bound
+    /// is not part of the property: a write that was queued instead of rejected can only
+    /// be answered after the gate opens, which never happens before the bound expires.
+    /// </remarks>
     [Theory]
     [InlineData(false)]
     [InlineData(true)]
@@ -170,25 +180,35 @@ public class HighReviewRegressionTests : TestKit
     {
         var adapter = Adapter();
         var gate = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
-        if (!reconnect) adapter.ConnectAsync(Arg.Any<IDictionary<string, string>>(), Arg.Any<CancellationToken>()).Returns(gate.Task);
+        var attemptIssued = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
+        void HoldConnectAttempts() => adapter.ConnectAsync(Arg.Any<IDictionary<string, string>>(), Arg.Any<CancellationToken>())
+            .Returns(_ => { attemptIssued.TrySetResult(); return gate.Task; });
+        if (!reconnect) HoldConnectAttempts();
         var actor = Actor(adapter);
         if (reconnect)
         {
-            Connected();
-            adapter.ConnectAsync(Arg.Any<IDictionary<string, string>>(), Arg.Any<CancellationToken>()).Returns(gate.Task);
+            AwaitCondition(() => HealthReported(ConnectionHealth.Connected), BarrierTimeout);
+            HoldConnectAttempts();
             adapter.Disconnected += Raise.Event<Action>();
-            AwaitCondition(() => _health.ReceivedCalls().Any(c => c.GetMethodInfo().Name == "UpdateConnectionHealth" && c.GetArguments()[1] is ConnectionHealth.Disconnected));
+            AwaitCondition(() => HealthReported(ConnectionHealth.Disconnected), BarrierTimeout);
         }
+        AwaitCondition(() => attemptIssued.Task.IsCompleted, BarrierTimeout);
         actor.Tell(new WriteTagRequest("single", "conn", "tag", 42, DateTimeOffset.UtcNow));
-        ExpectMsg<WriteTagResponse>(r => !r.Success && r.CorrelationId == "single", TimeSpan.FromSeconds(1));
+        ExpectMsg<WriteTagResponse>(r => !r.Success && r.CorrelationId == "single", BarrierTimeout);
         actor.Tell(new WriteTagBatchRequest("batch", "conn", new Dictionary<string, object?> { ["tag"] = 42 }, null, null, DateTimeOffset.UtcNow));
-        ExpectMsg<WriteTagBatchResponse>(r => !r.Success && r.CorrelationId == "batch", TimeSpan.FromSeconds(1));
+        ExpectMsg<WriteTagBatchResponse>(r => !r.Success && r.CorrelationId == "batch", BarrierTimeout);
         gate.SetResult();
         actor.Tell(Subscribe());
-        ExpectMsg<SubscribeTagsResponse>();
+        ExpectMsg<SubscribeTagsResponse>(BarrierTimeout);
         Assert.DoesNotContain(adapter.ReceivedCalls(), c => c.GetMethodInfo().Name is "WriteAsync" or "WriteBatchAsync");
     }
 
+    // Upper bound for barrier waits; never a property under test (see ScadaBridge#101).
+    private static readonly TimeSpan BarrierTimeout = TimeSpan.FromSeconds(10);
+
+    private bool HealthReported(ConnectionHealth state) => _health.ReceivedCalls().Any(c =>
+        c.GetMethodInfo().Name == "UpdateConnectionHealth" && c.GetArguments()[1] is ConnectionHealth h && h == state);
+
     /// <summary>Verifies that a subscriber terminating releases the handle of its last tag.</summary>
     [Fact]
     public void SubscriberTermination_ReleasesLastTagHandle()