Phase 3 PR 53 -- Transport reconnect-on-drop + SO_KEEPALIVE for DL205 no-keepalive quirk. AutomationDirect H2-ECOM100 does NOT send TCP keepalives per docs/v2/dl205.md behavioral-oddities section -- any NAT/firewall device between the gateway and the PLC can silently close an idle socket after 2-5 minutes of inactivity. The PLC itself never notices and the first SendAsync after the drop would previously surface as IOException / EndOfStreamException / SocketException to the caller even though the PLC is perfectly healthy. PR 53 makes ModbusTcpTransport survive mid-session socket drops: SendAsync wraps the previous body as SendOnceAsync; on the first attempt, if the failure is a socket-layer error (IOException, SocketException, EndOfStreamException, ObjectDisposedException) AND autoReconnect is enabled (default true), the transport tears down the dead socket, calls ConnectAsync to re-establish, and resends the PDU exactly once. Deliberately single-retry -- further failures propagate so the driver health surface reflects the real state, no masking a dead PLC. Protocol-layer failures (e.g. ModbusException with exception code 02) are specifically NOT caught by the reconnect path -- they would just come back with the same exception code after the reconnect, so retrying is wasted wire time. Socket-level vs protocol-level is a discriminator inside IsSocketLevelFailure. Also enables SO_KEEPALIVE on the TcpClient with aggressive timing: TcpKeepAliveTime=30s, TcpKeepAliveInterval=10s, TcpKeepAliveRetryCount=3. Total time-to-detect-dead-socket = 30 + 10*3 = 60s, vs the Windows default 2-hour idle + 9 retries = 2h40min. Best-effort: older OSes that don't expose the fine-grained keepalive knobs silently skip them (catch {}). New ModbusDriverOptions.AutoReconnect bool (default true) threads through to the default transport factory in ModbusDriver -- callers wanting the old 'fail loud on drop' behavior can set AutoReconnect=false, or use a custom transportFactory that ignores the option. Unit tests: ModbusTcpReconnectTests boots a FlakeyModbusServer in-process (real TcpListener on loopback) that serves one valid FC03 response then forcibly shuts down the socket. Transport_recovers_from_mid_session_drop_and_retries_successfully issues two consecutive SendAsync calls and asserts both return valid PDUs -- the second must trigger the reconnect path transparently. Transport_without_AutoReconnect_propagates_drop_to_caller asserts the legacy behavior when the opt-out is taken. Validates real socket semantics rather than mocked exceptions. 142/142 Modbus.Tests pass (113 prior + 2 mapper + 2 reconnect + 25 accumulated across PRs 45-52); 11/11 DL205 integration tests still pass with MODBUS_SIM_PROFILE=dl205 -- no regression from the transport change.

src/ZB.MOM.WW.OtOpcUa.Driver.Modbus/ModbusDriver.cs

@@ -37,7 +37,7 @@ public sealed class ModbusDriver(ModbusDriverOptions options, string driverInsta
     private CancellationTokenSource? _probeCts;
     private readonly ModbusDriverOptions _options = options;
     private readonly Func<ModbusDriverOptions, IModbusTransport> _transportFactory =
-        transportFactory ?? (o => new ModbusTcpTransport(o.Host, o.Port, o.Timeout));
+        transportFactory ?? (o => new ModbusTcpTransport(o.Host, o.Port, o.Timeout, o.AutoReconnect));
 
     private IModbusTransport? _transport;
     private DriverHealth _health = new(DriverState.Unknown, null, null);
@@ -141,9 +141,16 @@ public sealed class ModbusDriver(ModbusDriverOptions options, string driverInsta
                 results[i] = new DataValueSnapshot(value, 0u, now, now);
                 _health = new DriverHealth(DriverState.Healthy, now, null);
             }
+            catch (ModbusException mex)
+            {
+                results[i] = new DataValueSnapshot(null, MapModbusExceptionToStatus(mex.ExceptionCode), null, now);
+                _health = new DriverHealth(DriverState.Degraded, _health.LastSuccessfulRead, mex.Message);
+            }
             catch (Exception ex)
             {
-                results[i] = new DataValueSnapshot(null, StatusBadInternalError, null, now);
+                // Non-Modbus-layer failure: socket dropped, timeout, malformed response. Surface
+                // as communication error so callers can distinguish it from tag-level faults.
+                results[i] = new DataValueSnapshot(null, StatusBadCommunicationError, null, now);
                 _health = new DriverHealth(DriverState.Degraded, _health.LastSuccessfulRead, ex.Message);
             }
         }
@@ -238,6 +245,10 @@ public sealed class ModbusDriver(ModbusDriverOptions options, string driverInsta
                 await WriteOneAsync(transport, tag, w.Value, cancellationToken).ConfigureAwait(false);
                 results[i] = new WriteResult(0u);
             }
+            catch (ModbusException mex)
+            {
+                results[i] = new WriteResult(MapModbusExceptionToStatus(mex.ExceptionCode));
+            }
             catch (Exception)
             {
                 results[i] = new WriteResult(StatusBadInternalError);
@@ -686,6 +697,31 @@ public sealed class ModbusDriver(ModbusDriverOptions options, string driverInsta
     private const uint StatusBadInternalError = 0x80020000u;
     private const uint StatusBadNodeIdUnknown = 0x80340000u;
     private const uint StatusBadNotWritable = 0x803B0000u;
+    private const uint StatusBadOutOfRange = 0x803C0000u;
+    private const uint StatusBadNotSupported = 0x803D0000u;
+    private const uint StatusBadDeviceFailure = 0x80550000u;
+    private const uint StatusBadCommunicationError = 0x80050000u;
+
+    /// <summary>
+    ///     Map a server-returned Modbus exception code to the most informative OPC UA
+    ///     StatusCode. Keeps the driver's outward-facing status surface aligned with what a
+    ///     Modbus engineer would expect when reading the spec: exception 02 (Illegal Data
+    ///     Address) surfaces as BadOutOfRange so clients can distinguish "tag wrong" from
+    ///     generic BadInternalError, exception 04 (Server Failure) as BadDeviceFailure so
+    ///     operators see a CPU-mode problem rather than a driver bug, etc. Per
+    ///     <c>docs/v2/dl205.md</c>, DL205/DL260 returns only codes 01-04 — no proprietary
+    ///     extensions.
+    /// </summary>
+    internal static uint MapModbusExceptionToStatus(byte exceptionCode) => exceptionCode switch
+    {
+        0x01 => StatusBadNotSupported,     // Illegal Function — FC not in supported list
+        0x02 => StatusBadOutOfRange,       // Illegal Data Address — register outside mapped range
+        0x03 => StatusBadOutOfRange,       // Illegal Data Value — quantity over per-FC cap
+        0x04 => StatusBadDeviceFailure,    // Server Failure — CPU in PROGRAM mode during protected write
+        0x05 or 0x06 => StatusBadDeviceFailure, // Acknowledge / Server Busy — long-running op / busy
+        0x0A or 0x0B => StatusBadCommunicationError, // Gateway path unavailable / target failed to respond
+        _ => StatusBadInternalError,
+    };
 
     public void Dispose() => DisposeAsync().AsTask().GetAwaiter().GetResult();
     public async ValueTask DisposeAsync()

src/ZB.MOM.WW.OtOpcUa.Driver.Modbus/ModbusDriverOptions.cs

@@ -45,6 +45,17 @@ public sealed class ModbusDriverOptions
     ///     by shortening the tag's <c>StringLength</c> or splitting it into multiple tags).
     /// </summary>
     public ushort MaxRegistersPerWrite { get; init; } = 123;
+
+    /// <summary>
+    ///     When <c>true</c> (default) the built-in <see cref="ModbusTcpTransport"/> detects
+    ///     mid-transaction socket failures (<see cref="System.IO.EndOfStreamException"/>,
+    ///     <see cref="System.Net.Sockets.SocketException"/>) and transparently reconnects +
+    ///     retries the PDU exactly once. Required for DL205/DL260 because the H2-ECOM100
+    ///     does not send TCP keepalives — intermediate NAT / firewall devices silently close
+    ///     idle sockets and the first send after the drop would otherwise surface as a
+    ///     connection error to the caller even though the PLC is up.
+    /// </summary>
+    public bool AutoReconnect { get; init; } = true;
 }
 
 public sealed class ModbusProbeOptions

src/ZB.MOM.WW.OtOpcUa.Driver.Modbus/ModbusTcpTransport.cs

@@ -8,22 +8,40 @@ namespace ZB.MOM.WW.OtOpcUa.Driver.Modbus;
 ///     support concurrent transactions, but the single-flight model keeps the wire trace
 ///     easy to diagnose and avoids interleaved-response correlation bugs.
 /// </summary>
+/// <remarks>
+///     <para>
+///         Survives mid-transaction socket drops: when a send/read fails with a socket-level
+///         error (<see cref="IOException"/>, <see cref="SocketException"/>, <see cref="EndOfStreamException"/>)
+///         the transport disposes the dead socket, reconnects, and retries the PDU exactly
+///         once. Deliberately limited to a single retry — further failures bubble up so the
+///         driver's health surface reflects the real state instead of masking a dead PLC.
+///     </para>
+///     <para>
+///         Why this matters for DL205/DL260: the AutomationDirect H2-ECOM100 does NOT send
+///         TCP keepalives per <c>docs/v2/dl205.md</c> §behavioral-oddities, so any NAT/firewall
+///         between the gateway and PLC can silently close an idle socket after 2-5 minutes.
+///         Also enables OS-level <c>SO_KEEPALIVE</c> so the driver's own side detects a stuck
+///         socket in reasonable time even when the application is mostly idle.
+///     </para>
+/// </remarks>
 public sealed class ModbusTcpTransport : IModbusTransport
 {
     private readonly string _host;
     private readonly int _port;
     private readonly TimeSpan _timeout;
+    private readonly bool _autoReconnect;
     private readonly SemaphoreSlim _gate = new(1, 1);
     private TcpClient? _client;
     private NetworkStream? _stream;
     private ushort _nextTx;
     private bool _disposed;
 
-    public ModbusTcpTransport(string host, int port, TimeSpan timeout)
+    public ModbusTcpTransport(string host, int port, TimeSpan timeout, bool autoReconnect = true)
     {
         _host = host;
         _port = port;
         _timeout = timeout;
+        _autoReconnect = autoReconnect;
     }
 
     public async Task ConnectAsync(CancellationToken ct)
@@ -39,12 +57,34 @@ public sealed class ModbusTcpTransport : IModbusTransport
         var target = ipv4 ?? (addresses.Length > 0 ? addresses[0] : System.Net.IPAddress.Loopback);
 
         _client = new TcpClient(target.AddressFamily);
+        EnableKeepAlive(_client);
+
         using var cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
         cts.CancelAfter(_timeout);
         await _client.ConnectAsync(target, _port, cts.Token).ConfigureAwait(false);
         _stream = _client.GetStream();
     }
 
+    /// <summary>
+    ///     Enable SO_KEEPALIVE with aggressive probe timing. DL205/DL260 doesn't send keepalives
+    ///     itself; having the OS probe the socket every ~30s lets the driver notice a dead PLC
+    ///     or broken NAT path long before the default 2-hour Windows idle timeout fires.
+    ///     Non-fatal if the underlying OS rejects the option (some older Linux / container
+    ///     sandboxes don't expose the fine-grained timing levers — the driver still works,
+    ///     application-level probe still detects problems).
+    /// </summary>
+    private static void EnableKeepAlive(TcpClient client)
+    {
+        try
+        {
+            client.Client.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.KeepAlive, true);
+            client.Client.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveTime, 30);
+            client.Client.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveInterval, 10);
+            client.Client.SetSocketOption(SocketOptionLevel.Tcp, SocketOptionName.TcpKeepAliveRetryCount, 3);
+        }
+        catch { /* best-effort; older OSes may not expose the granular knobs */ }
+    }
+
     public async Task<byte[]> SendAsync(byte unitId, byte[] pdu, CancellationToken ct)
     {
         if (_disposed) throw new ObjectDisposedException(nameof(ModbusTcpTransport));
@@ -53,43 +93,18 @@ public sealed class ModbusTcpTransport : IModbusTransport
         await _gate.WaitAsync(ct).ConfigureAwait(false);
         try
         {
-            var txId = ++_nextTx;
-
-            // MBAP: [TxId(2)][Proto=0(2)][Length(2)][UnitId(1)] + PDU
-            var adu = new byte[7 + pdu.Length];
-            adu[0] = (byte)(txId >> 8);
-            adu[1] = (byte)(txId & 0xFF);
-            // protocol id already zero
-            var len = (ushort)(1 + pdu.Length); // unit id + pdu
-            adu[4] = (byte)(len >> 8);
-            adu[5] = (byte)(len & 0xFF);
-            adu[6] = unitId;
-            Buffer.BlockCopy(pdu, 0, adu, 7, pdu.Length);
-
-            using var cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
-            cts.CancelAfter(_timeout);
-            await _stream.WriteAsync(adu.AsMemory(), cts.Token).ConfigureAwait(false);
-            await _stream.FlushAsync(cts.Token).ConfigureAwait(false);
-
-            var header = new byte[7];
-            await ReadExactlyAsync(_stream, header, cts.Token).ConfigureAwait(false);
-            var respTxId = (ushort)((header[0] << 8) | header[1]);
-            if (respTxId != txId)
-                throw new InvalidDataException($"Modbus TxId mismatch: expected {txId} got {respTxId}");
-            var respLen = (ushort)((header[4] << 8) | header[5]);
-            if (respLen < 1) throw new InvalidDataException($"Modbus response length too small: {respLen}");
-            var respPdu = new byte[respLen - 1];
-            await ReadExactlyAsync(_stream, respPdu, cts.Token).ConfigureAwait(false);
-
-            // Exception PDU: function code has high bit set.
-            if ((respPdu[0] & 0x80) != 0)
+            try
             {
-                var fc = (byte)(respPdu[0] & 0x7F);
-                var ex = respPdu[1];
-                throw new ModbusException(fc, ex, $"Modbus exception fc={fc} code={ex}");
+                return await SendOnceAsync(unitId, pdu, ct).ConfigureAwait(false);
+            }
+            catch (Exception ex) when (_autoReconnect && IsSocketLevelFailure(ex))
+            {
+                // Mid-transaction drop: tear down the dead socket, reconnect, resend. Single
+                // retry — if it fails again, let it propagate so health/status reflect reality.
+                await TearDownAsync().ConfigureAwait(false);
+                await ConnectAsync(ct).ConfigureAwait(false);
+                return await SendOnceAsync(unitId, pdu, ct).ConfigureAwait(false);
             }
-
-            return respPdu;
         }
         finally
         {
@@ -97,6 +112,68 @@ public sealed class ModbusTcpTransport : IModbusTransport
         }
     }
 
+    private async Task<byte[]> SendOnceAsync(byte unitId, byte[] pdu, CancellationToken ct)
+    {
+        if (_stream is null) throw new InvalidOperationException("Transport not connected");
+        var txId = ++_nextTx;
+
+        // MBAP: [TxId(2)][Proto=0(2)][Length(2)][UnitId(1)] + PDU
+        var adu = new byte[7 + pdu.Length];
+        adu[0] = (byte)(txId >> 8);
+        adu[1] = (byte)(txId & 0xFF);
+        // protocol id already zero
+        var len = (ushort)(1 + pdu.Length); // unit id + pdu
+        adu[4] = (byte)(len >> 8);
+        adu[5] = (byte)(len & 0xFF);
+        adu[6] = unitId;
+        Buffer.BlockCopy(pdu, 0, adu, 7, pdu.Length);
+
+        using var cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
+        cts.CancelAfter(_timeout);
+        await _stream.WriteAsync(adu.AsMemory(), cts.Token).ConfigureAwait(false);
+        await _stream.FlushAsync(cts.Token).ConfigureAwait(false);
+
+        var header = new byte[7];
+        await ReadExactlyAsync(_stream, header, cts.Token).ConfigureAwait(false);
+        var respTxId = (ushort)((header[0] << 8) | header[1]);
+        if (respTxId != txId)
+            throw new InvalidDataException($"Modbus TxId mismatch: expected {txId} got {respTxId}");
+        var respLen = (ushort)((header[4] << 8) | header[5]);
+        if (respLen < 1) throw new InvalidDataException($"Modbus response length too small: {respLen}");
+        var respPdu = new byte[respLen - 1];
+        await ReadExactlyAsync(_stream, respPdu, cts.Token).ConfigureAwait(false);
+
+        // Exception PDU: function code has high bit set.
+        if ((respPdu[0] & 0x80) != 0)
+        {
+            var fc = (byte)(respPdu[0] & 0x7F);
+            var ex = respPdu[1];
+            throw new ModbusException(fc, ex, $"Modbus exception fc={fc} code={ex}");
+        }
+
+        return respPdu;
+    }
+
+    /// <summary>
+    ///     Distinguish socket-layer failures (eligible for reconnect-and-retry) from
+    ///     protocol-layer failures (must propagate — retrying the same PDU won't help if the
+    ///     PLC just returned exception 02 Illegal Data Address).
+    /// </summary>
+    private static bool IsSocketLevelFailure(Exception ex) =>
+        ex is EndOfStreamException
+        || ex is IOException
+        || ex is SocketException
+        || ex is ObjectDisposedException;
+
+    private async Task TearDownAsync()
+    {
+        try { if (_stream is not null) await _stream.DisposeAsync().ConfigureAwait(false); }
+        catch { /* best-effort */ }
+        _stream = null;
+        try { _client?.Dispose(); } catch { }
+        _client = null;
+    }
+
     private static async Task ReadExactlyAsync(Stream s, byte[] buf, CancellationToken ct)
     {
         var read = 0;

tests/ZB.MOM.WW.OtOpcUa.Driver.Modbus.IntegrationTests/DL205/DL205ExceptionCodeTests.cs

@@ -0,0 +1,53 @@
+using Shouldly;
+using Xunit;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.Modbus.IntegrationTests.DL205;
+
+/// <summary>
+///     Verifies the driver's Modbus-exception → OPC UA StatusCode translation end-to-end
+///     against the dl205.json pymodbus profile. pymodbus returns exception 02 (Illegal Data
+///     Address) for reads outside the configured register ranges, matching real DL205/DL260
+///     firmware behavior per <c>docs/v2/dl205.md</c> §exception-codes. The driver must surface
+///     that as <c>BadOutOfRange</c> (0x803C0000) — not <c>BadInternalError</c> — so the
+///     operator sees a tag-config diagnosis instead of a generic driver-fault message.
+/// </summary>
+[Collection(ModbusSimulatorCollection.Name)]
+[Trait("Category", "Integration")]
+[Trait("Device", "DL205")]
+public sealed class DL205ExceptionCodeTests(ModbusSimulatorFixture sim)
+{
+    [Fact]
+    public async Task DL205_FC03_at_unmapped_register_returns_BadOutOfRange()
+    {
+        if (sim.SkipReason is not null) Assert.Skip(sim.SkipReason);
+        if (!string.Equals(Environment.GetEnvironmentVariable("MODBUS_SIM_PROFILE"), "dl205",
+                StringComparison.OrdinalIgnoreCase))
+        {
+            Assert.Skip("MODBUS_SIM_PROFILE != dl205 — skipping.");
+        }
+
+        // Address 16383 is the last cell of hr-size=16384 in dl205.json; address 16384 is
+        // beyond the configured HR range. pymodbus validates and returns exception 02
+        // (Illegal Data Address).
+        var options = new ModbusDriverOptions
+        {
+            Host = sim.Host,
+            Port = sim.Port,
+            UnitId = 1,
+            Timeout = TimeSpan.FromSeconds(2),
+            Tags =
+            [
+                new ModbusTagDefinition("Unmapped",
+                    ModbusRegion.HoldingRegisters, Address: 16383,
+                    DataType: ModbusDataType.UInt16, Writable: false),
+            ],
+            Probe = new ModbusProbeOptions { Enabled = false },
+        };
+        await using var driver = new ModbusDriver(options, driverInstanceId: "dl205-exc");
+        await driver.InitializeAsync("{}", TestContext.Current.CancellationToken);
+
+        var results = await driver.ReadAsync(["Unmapped"], TestContext.Current.CancellationToken);
+        results[0].StatusCode.ShouldBe(0x803C0000u,
+            "DL205 returns exception 02 for an FC03 at an unmapped register; driver must translate to BadOutOfRange (not BadInternalError)");
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.Modbus.Tests/ModbusExceptionMapperTests.cs

@@ -0,0 +1,88 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.Modbus.Tests;
+
+/// <summary>
+///     Unit tests for the Modbus-exception-code → OPC UA StatusCode mapping added in PR 52.
+///     Before PR 52 every server exception + every transport failure collapsed to
+///     BadInternalError (0x80020000), which made field diagnosis "is this a bad tag or a bad
+///     driver?" impossible. These tests lock in the translation table documented on
+///     <see cref="ModbusDriver.MapModbusExceptionToStatus"/>.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class ModbusExceptionMapperTests
+{
+    [Theory]
+    [InlineData((byte)0x01, 0x803D0000u)] // Illegal Function → BadNotSupported
+    [InlineData((byte)0x02, 0x803C0000u)] // Illegal Data Address → BadOutOfRange
+    [InlineData((byte)0x03, 0x803C0000u)] // Illegal Data Value → BadOutOfRange
+    [InlineData((byte)0x04, 0x80550000u)] // Server Failure → BadDeviceFailure
+    [InlineData((byte)0x05, 0x80550000u)] // Acknowledge (long op) → BadDeviceFailure
+    [InlineData((byte)0x06, 0x80550000u)] // Server Busy → BadDeviceFailure
+    [InlineData((byte)0x0A, 0x80050000u)] // Gateway path unavailable → BadCommunicationError
+    [InlineData((byte)0x0B, 0x80050000u)] // Gateway target failed to respond → BadCommunicationError
+    [InlineData((byte)0xFF, 0x80020000u)] // Unknown code → BadInternalError fallback
+    public void MapModbusExceptionToStatus_returns_informative_status(byte code, uint expected)
+        => ModbusDriver.MapModbusExceptionToStatus(code).ShouldBe(expected);
+
+    private sealed class ExceptionRaisingTransport(byte exceptionCode) : IModbusTransport
+    {
+        public Task ConnectAsync(CancellationToken ct) => Task.CompletedTask;
+        public Task<byte[]> SendAsync(byte unitId, byte[] pdu, CancellationToken ct)
+            => Task.FromException<byte[]>(new ModbusException(pdu[0], exceptionCode, $"fc={pdu[0]} code={exceptionCode}"));
+        public ValueTask DisposeAsync() => ValueTask.CompletedTask;
+    }
+
+    [Fact]
+    public async Task Read_surface_exception_02_as_BadOutOfRange_not_BadInternalError()
+    {
+        var transport = new ExceptionRaisingTransport(exceptionCode: 0x02);
+        var tag = new ModbusTagDefinition("T", ModbusRegion.HoldingRegisters, 0, ModbusDataType.Int16);
+        var opts = new ModbusDriverOptions { Host = "fake", Tags = [tag], Probe = new ModbusProbeOptions { Enabled = false } };
+        await using var drv = new ModbusDriver(opts, "modbus-1", _ => transport);
+        await drv.InitializeAsync("{}", TestContext.Current.CancellationToken);
+
+        var results = await drv.ReadAsync(["T"], TestContext.Current.CancellationToken);
+        results[0].StatusCode.ShouldBe(0x803C0000u, "FC03 at an unmapped register must bubble out as BadOutOfRange so operators can spot a bad tag config");
+    }
+
+    [Fact]
+    public async Task Write_surface_exception_04_as_BadDeviceFailure()
+    {
+        var transport = new ExceptionRaisingTransport(exceptionCode: 0x04);
+        var tag = new ModbusTagDefinition("T", ModbusRegion.HoldingRegisters, 0, ModbusDataType.Int16);
+        var opts = new ModbusDriverOptions { Host = "fake", Tags = [tag], Probe = new ModbusProbeOptions { Enabled = false } };
+        await using var drv = new ModbusDriver(opts, "modbus-1", _ => transport);
+        await drv.InitializeAsync("{}", TestContext.Current.CancellationToken);
+
+        var writes = await drv.WriteAsync(
+            [new WriteRequest("T", (short)42)],
+            TestContext.Current.CancellationToken);
+
+        writes[0].StatusCode.ShouldBe(0x80550000u, "FC06 returning exception 04 (CPU in PROGRAM mode) maps to BadDeviceFailure");
+    }
+
+    private sealed class NonModbusFailureTransport : IModbusTransport
+    {
+        public Task ConnectAsync(CancellationToken ct) => Task.CompletedTask;
+        public Task<byte[]> SendAsync(byte unitId, byte[] pdu, CancellationToken ct)
+            => Task.FromException<byte[]>(new EndOfStreamException("socket closed mid-response"));
+        public ValueTask DisposeAsync() => ValueTask.CompletedTask;
+    }
+
+    [Fact]
+    public async Task Read_non_modbus_failure_maps_to_BadCommunicationError_not_BadInternalError()
+    {
+        // Socket drop / timeout / malformed frame → transport-layer failure. Should surface
+        // distinctly from tag-level faults so operators know to check the network, not the config.
+        var tag = new ModbusTagDefinition("T", ModbusRegion.HoldingRegisters, 0, ModbusDataType.Int16);
+        var opts = new ModbusDriverOptions { Host = "fake", Tags = [tag], Probe = new ModbusProbeOptions { Enabled = false } };
+        await using var drv = new ModbusDriver(opts, "modbus-1", _ => new NonModbusFailureTransport());
+        await drv.InitializeAsync("{}", TestContext.Current.CancellationToken);
+
+        var results = await drv.ReadAsync(["T"], TestContext.Current.CancellationToken);
+        results[0].StatusCode.ShouldBe(0x80050000u);
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.Modbus.Tests/ModbusTcpReconnectTests.cs

@@ -0,0 +1,146 @@
+using System.Net;
+using System.Net.Sockets;
+using Shouldly;
+using Xunit;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.Modbus.Tests;
+
+/// <summary>
+///     Exercises <see cref="ModbusTcpTransport"/> against a real TCP listener that can close
+///     its socket mid-session on demand. Verifies the PR 53 reconnect-on-drop behavior: after
+///     the "first" socket is forcibly torn down, the next SendAsync must re-establish the
+///     connection and complete the PDU without bubbling an error to the caller.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class ModbusTcpReconnectTests
+{
+    /// <summary>
+    ///     Minimal in-process Modbus-TCP stub. Accepts one TCP connection at a time, reads an
+    ///     MBAP + PDU, replies with a canned FC03 response echoing the request quantity of
+    ///     zeroed bytes, then optionally closes the socket to simulate a NAT/firewall drop.
+    /// </summary>
+    private sealed class FlakeyModbusServer : IAsyncDisposable
+    {
+        private readonly TcpListener _listener;
+        public int Port => ((IPEndPoint)_listener.LocalEndpoint).Port;
+        public int DropAfterNTransactions { get; set; } = int.MaxValue;
+        private readonly CancellationTokenSource _stop = new();
+        private int _txCount;
+
+        public FlakeyModbusServer()
+        {
+            _listener = new TcpListener(IPAddress.Loopback, 0);
+            _listener.Start();
+            _ = Task.Run(AcceptLoopAsync);
+        }
+
+        private async Task AcceptLoopAsync()
+        {
+            while (!_stop.IsCancellationRequested)
+            {
+                TcpClient? client = null;
+                try { client = await _listener.AcceptTcpClientAsync(_stop.Token); }
+                catch { return; }
+
+                _ = Task.Run(() => ServeAsync(client!));
+            }
+        }
+
+        private async Task ServeAsync(TcpClient client)
+        {
+            try
+            {
+                using var _ = client;
+                var stream = client.GetStream();
+                while (!_stop.IsCancellationRequested && client.Connected)
+                {
+                    var header = new byte[7];
+                    if (!await ReadExactly(stream, header)) return;
+                    var len = (ushort)((header[4] << 8) | header[5]);
+                    var pdu = new byte[len - 1];
+                    if (!await ReadExactly(stream, pdu)) return;
+
+                    var fc = pdu[0];
+                    var qty = (ushort)((pdu[3] << 8) | pdu[4]);
+                    var respPdu = new byte[2 + qty * 2];
+                    respPdu[0] = fc;
+                    respPdu[1] = (byte)(qty * 2);
+                    // data bytes stay 0
+
+                    var respLen = (ushort)(1 + respPdu.Length);
+                    var adu = new byte[7 + respPdu.Length];
+                    adu[0] = header[0]; adu[1] = header[1];
+                    adu[4] = (byte)(respLen >> 8); adu[5] = (byte)(respLen & 0xFF);
+                    adu[6] = header[6];
+                    Buffer.BlockCopy(respPdu, 0, adu, 7, respPdu.Length);
+                    await stream.WriteAsync(adu);
+                    await stream.FlushAsync();
+
+                    _txCount++;
+                    if (_txCount >= DropAfterNTransactions)
+                    {
+                        // Simulate NAT/firewall silent close: slam the socket without a
+                        // protocol-level goodbye, which is what DL260 + an intermediate
+                        // middlebox would look like from the client's perspective.
+                        client.Client.Shutdown(SocketShutdown.Both);
+                        client.Close();
+                        return;
+                    }
+                }
+            }
+            catch { /* best-effort */ }
+        }
+
+        private static async Task<bool> ReadExactly(NetworkStream s, byte[] buf)
+        {
+            var read = 0;
+            while (read < buf.Length)
+            {
+                var n = await s.ReadAsync(buf.AsMemory(read));
+                if (n == 0) return false;
+                read += n;
+            }
+            return true;
+        }
+
+        public async ValueTask DisposeAsync()
+        {
+            _stop.Cancel();
+            _listener.Stop();
+            await Task.CompletedTask;
+        }
+    }
+
+    [Fact]
+    public async Task Transport_recovers_from_mid_session_drop_and_retries_successfully()
+    {
+        await using var server = new FlakeyModbusServer { DropAfterNTransactions = 1 };
+        await using var transport = new ModbusTcpTransport("127.0.0.1", server.Port, TimeSpan.FromSeconds(2), autoReconnect: true);
+        await transport.ConnectAsync(TestContext.Current.CancellationToken);
+
+        // First transaction succeeds; server then closes the socket.
+        var pdu = new byte[] { 0x03, 0x00, 0x00, 0x00, 0x01 };
+        var first = await transport.SendAsync(unitId: 1, pdu, TestContext.Current.CancellationToken);
+        first[0].ShouldBe((byte)0x03);
+
+        // Second transaction: the connection is dead, but auto-reconnect must transparently
+        // spin up a new socket, resend, and produce a valid response. Before PR 53 this would
+        // surface as EndOfStreamException / IOException to the caller.
+        var second = await transport.SendAsync(unitId: 1, pdu, TestContext.Current.CancellationToken);
+        second[0].ShouldBe((byte)0x03);
+    }
+
+    [Fact]
+    public async Task Transport_without_AutoReconnect_propagates_drop_to_caller()
+    {
+        await using var server = new FlakeyModbusServer { DropAfterNTransactions = 1 };
+        await using var transport = new ModbusTcpTransport("127.0.0.1", server.Port, TimeSpan.FromSeconds(2), autoReconnect: false);
+        await transport.ConnectAsync(TestContext.Current.CancellationToken);
+
+        var pdu = new byte[] { 0x03, 0x00, 0x00, 0x00, 0x01 };
+        _ = await transport.SendAsync(unitId: 1, pdu, TestContext.Current.CancellationToken);
+
+        await Should.ThrowAsync<Exception>(async () =>
+            await transport.SendAsync(unitId: 1, pdu, TestContext.Current.CancellationToken));
+    }
+}