Merge review R1 fixes: removed driver's resilience tracker released; SDK timed-unshelve no-op (engine owns expiry); commented re-ack stored as AddComment; native-ack Primary verdict published from start; drain gate SharesRowsWithPeer; malformed cache bytes don't adopt the revision; HostAbortedException passthrough; ClusterRoleInfo seed filtered

Claude-Session: https://claude.ai/code/session_01Ef2ozxS7QPanGT8uoATbTe
2026-10-08 07:47:57 -04:00
parent b7a48e2b93 e3dcb180c3
commit b99aea46b2
28 changed files with 691 additions and 136 deletions
@@ -93,6 +93,16 @@ Two postures are deliberate:
 - **Unset (the default) means drain.** So does an unpublished view. A deployment
  that runs no redundancy never publishes a role, and defaulting closed would
  silently stop its alarm history forever.
 - **Only a shared queue is gated.** Standing down is safe only because a peer
  holds the same rows, so the gate applies only when
  `LocalDbReplicationPolicy.SharesRowsWithPeer` holds:
  `LocalDb:Replication:ApiKey` **AND** (`LocalDb:Replication:PeerAddress` **OR**
  `LocalDb:SyncListenPort > 0`) — an authenticated sync path in either
  direction. Any other node always drains its own queue. This is deliberately
  stricter than the table-registration predicate (`PeerAddress` OR `ApiKey`):
  a key with no dial and no listener (e.g. a fleet-wide secret on an unpaired
  node) captures changes but shares them with nobody, and gating it would park
  its events behind a Primary that does not hold them (review R1 m3).
 - **A gate that throws means "not now", never permission.** Draining on a gate
  fault would put a pair into dual delivery exactly when its redundancy signal is
  sick.
@@ -350,15 +350,16 @@ is client-visible.
 > (ScadaBridge included) that acks against the Secondary now gets a Bad result
 > instead of a Good that later reverts — so ack against the Primary (the higher
 > ServiceLevel). Scripted conditions are **not** gated: their engine runs on
-> both nodes, so an ack on the Secondary is real. Residual: the published
+> both nodes, so an ack on the Secondary is real. The published verdict
-> verdict changes only on a redundancy snapshot, so in the boot window of a
+> mirrors `DriverHostActor`'s live check from the moment the host starts, not
-> multi-node cluster (no snapshot yet) the node still accepts, while
+> only from the first redundancy snapshot: the host publishes it once at start
-> `DriverHostActor`'s live check may drop the route — the old flap, limited to
+> and again whenever cluster membership changes, so in the boot window of a
-> that window.
+> pair (role still unknown, a driver peer Up) the gate already denies, exactly
 > as the route does (review R1 m2 — previously the view sat on its single-node
 > "allow" seed until the first snapshot, reopening the flap for that window).
 > **Best-effort during reconnect.** If a native ack arrives while the owning
-> driver is `Connecting`/`Reconnecting` (or, in the boot window above, on a
+> driver is `Connecting`/`Reconnecting`, it is dropped with a debug log and **not**
 > node whose role is not yet known), it is dropped with a debug log and **not**
 > retried upstream — the OPC UA Part 9 `AlarmConditionState` was already
 > committed locally at step 1, so the rare outcome is "acked locally but not
 > propagated to AVEVA." Re-issue the ack once the driver is reconnected if
@@ -408,8 +409,10 @@ On allow, the handler publishes a `Commons.OpcUa.AlarmCommand` onto the
 action is a settable `Action<AlarmCommand>` injected at boot by the
 hosted service.
-`OnTimedUnshelve` (the SDK's automatic unshelve timer) bypasses the
+`OnTimedUnshelve` (the SDK's automatic unshelve timer) is a **no-op** — it
-operator gate — it is system-initiated.
+routes nothing (review R1 M2). The engine owns timed-shelve expiry (its 5 s
 shelving check records `AutoUnshelve` by `system`), and its re-projection
 unshelves the node; see §"`UnshelveTime` is the real remaining time" below.
 ### One condition event per method call (Gitea #587)
@@ -448,7 +451,14 @@ reports nothing.
  AddComment always reports (the comment is the change), and so does a re-ack
  **carrying a comment** (Gitea #597): the SDK writes that comment to the node,
  so it is reported once, with Acked/Confirmed unchanged. Before #597 the
-  comment landed on the node but no subscriber ever saw it.
+  comment landed on the node but no subscriber ever saw it. For a **scripted**
  condition the comment is also durable (review R1 m1): the routed Acknowledge
  reaches an engine that is already acknowledged, and
  `Part9StateMachine.ApplyAcknowledge` records a commented re-ack exactly as an
  AddComment (kind `AddComment`, the caller as user) — persisted and replicated
  with the condition state, the original ack's `LastAck*` fields untouched. An
  uncommented re-ack is still a no-op in the engine. The same applies to a
  commented re-ack from the AdminUI `/alerts` page.
 - **A native re-ack is still routed to the driver** (deliberate, #597). The
  route is fire-and-forget, so a node showing Acked does not prove the
  upstream system acknowledged — an ack dropped during a reconnect leaves
@@ -471,10 +481,14 @@ reports nothing.
  engine's projection carries its expiry (`AlarmConditionSnapshot.UnshelveAtUtc`),
  so `ShelvingState/UnshelveTime` counts down to the engine's expiry whichever
  path shelved it (the AdminUI `/alerts` timed shelve included). This arms the
-  SDK's own unshelve timer at the real instant; it changes no node state and
+  SDK's own unshelve timer at the real instant, and `OnTimedUnshelve` — the
-  only routes an `Unshelve` (user `system`) back to the engine — a no-op if the
+  timer's callback — is deliberately a **no-op** (review R1 M2): it used to
-  engine's 5 s expiry sweep got there first, otherwise it simply lands the
+  route an operator-style `Unshelve` by `system`, which, firing before the
-  expiry on time.
+  engine's 5 s expiry sweep, pre-empted the engine's `AutoUnshelve` on every
  expiry, from both pair nodes, rewriting the audit trail as a manual unshelve.
  The SDK's timer changes no shelving state itself, so the node stays
  TimedShelved (with `UnshelveTime` at zero) for up to one sweep after expiry,
  until the engine's re-projection reports the one expiry event.
 - **Accepted, not fixed — the SDK's `EventsMonitored` unit for late conditions**
  (Gitea #597 item 6, from the #586 review). The SDK's `AreEventsMonitored`
  counter never reaches a condition materialised under a folder that already had
@@ -218,7 +218,7 @@ See the [ClusterClient→gRPC migration design](plans/2026-08-06-clusterclient-t
 | `GrpcListenPort` | int | `0` | Dedicated **h2c-only** Kestrel port for the artifact gRPC service. `0` = disabled (nothing bound). Must differ from the main HTTP port and from `LocalDb:SyncListenPort` (h2c cannot share a cleartext HTTP/1 port). Mapped only on admin-role nodes. |
 | `ApiKey` | string | `""` | Shared bearer key the serve-side interceptor checks (constant-time, **fail-closed**: an unset key rejects every call). **Supply via env `ConfigServe__ApiKey` — never commit.** |
-**It is a dark switch; the serve side is always wired.** Central maps the `DeploymentArtifactService` behind the `ConfigServeAuthInterceptor` (path-scoped, `FixedTimeEquals`, fail-closed) whenever `ConfigServe:GrpcListenPort > 0`, in both modes — so flipping a node to `FetchAndCache` is a config change, not a redeploy. A `FetchAndCache` node that cannot fetch (all endpoints down, `NotFound`, SHA mismatch) **fails the apply and keeps last-known-good** — a null fetch is "no answer," never an empty configuration (the [#485](plans/2026-07-15-v3-batch4-address-space-plan.md) contract). At boot it restores served state from the LocalDb pointer with no SQL read; an empty or unreadable cache lands Steady-with-no-revision (the first dispatch fetches), never Stale.
+**It is a dark switch; the serve side is always wired.** Central maps the `DeploymentArtifactService` behind the `ConfigServeAuthInterceptor` (path-scoped, `FixedTimeEquals`, fail-closed) whenever `ConfigServe:GrpcListenPort > 0`, in both modes — so flipping a node to `FetchAndCache` is a config change, not a redeploy. A `FetchAndCache` node that cannot fetch (all endpoints down, `NotFound`, SHA mismatch) **fails the apply and keeps last-known-good** — a null fetch is "no answer," never an empty configuration (the [#485](plans/2026-07-15-v3-batch4-address-space-plan.md) contract). At boot it restores served state from the LocalDb pointer with no SQL read; an empty or unreadable cache — including a cached row whose bytes are malformed (review R1 m4) — lands Steady-with-no-revision (the first dispatch fetches, even of the same revision), never Stale.
 **The listener is dedicated h2c and coexists with the LocalDb-sync listener.** A fused admin+driver node with **both** `ConfigServe:GrpcListenPort` and `LocalDb:SyncListenPort` set re-binds its existing HTTP surface exactly once and adds both dedicated HTTP/2-only ports on top — see the same Kestrel-takeover caveat under [`LocalDb`](#localdb-pair-local-config-cache--driver-role-only).
@@ -431,7 +431,10 @@ replication stays inert until the sync port + peer are set.
 `PeerAddress` **OR** `ApiKey` non-blank — the OR because only the *dialling* half sets `PeerAddress`, so
 keying on it alone would strip change capture from every passive node and leave the pair converging in one
 direction only. `SyncListenPort` is deliberately **not** part of it (a port with no `ApiKey` is fail-closed
-and syncs with nobody). A configured node registers all four tables with `baselineExistingRows: true`, so
+and syncs with nobody). (The alarm-historian **drain gate** uses a stricter predicate on purpose —
 `ApiKey` AND (`PeerAddress` OR `SyncListenPort > 0`), `LocalDbReplicationPolicy.SharesRowsWithPeer` — because
 gating a node that shares its queue with nobody strands its alarm history; see `docs/AlarmHistorian.md`
 §"The drain gate".) A configured node registers all four tables with `baselineExistingRows: true`, so
 turning replication on converges the rows **already in the file** and not merely subsequent writes; an
 unconfigured node **deregisters** them at boot, dropping capture triggers an earlier build installed. ⚠️ The
 sync handshake compares registered-table digests **fail-closed**, so flipping either key is a
@@ -526,13 +526,13 @@ Three data-plane surfaces are **Primary-gated** so a warm standby never touches
 Before archreview 03/S4 the unknown-role case defaulted to *allow* unconditionally, opening a dual-primary window: a freshly-booted secondary on a multi-node cluster serviced shared-device writes and emitted duplicate fleet-wide alerts for up to the ~10 s `RedundancyStateActor` heartbeat (and indefinitely if the snapshot's NodeId never matched this node). Membership is authoritative much earlier than DPS delivery, so an unknown role on a **multi-driver** cluster now default-**denies**. The driver member count is read live from `Cluster.State.Members` (Up members with the `driver` role); a non-cluster ActorRefProvider yields 0 ⇒ single-node posture. If a snapshot arrives that never mentions this node while a driver peer exists (the 03/S5 identity-mismatch shape), `DriverHostActor` logs a one-time Warning so the silent mismatch is diagnosable.
-**What a client sees during the boot window (multi-node, role unknown):** an inbound write is **rejected** (not queued — OT actuation seconds late is a surprise). The reject rides the sanctioned optimistic-write self-correction surface: the node reverts to its prior value with a transient Bad-quality blip and a Part 8 `AuditWriteUpdateEvent`; the client's synchronous write call still returned Good (exactly as any failed device write behaves since the #5 write-outcome self-correction). The rejection reason distinguishes the boot window (`"not primary (role unknown)"`) from a steady-state Secondary (`"not primary"`). Acks are dropped (Warning-logged when role-unknown). The alerts emit is skipped (the Primary peer publishes the single fleet copy). The window closes on the first delivered snapshot.
+**What a client sees during the boot window (multi-node, role unknown):** an inbound write is **rejected** (not queued — OT actuation seconds late is a surprise). The reject rides the sanctioned optimistic-write self-correction surface: the node reverts to its prior value with a transient Bad-quality blip and a Part 8 `AuditWriteUpdateEvent`; the client's synchronous write call still returned Good (exactly as any failed device write behaves since the #5 write-outcome self-correction). The rejection reason distinguishes the boot window (`"not primary (role unknown)"`) from a steady-state Secondary (`"not primary"`). Acks are dropped (Warning-logged when role-unknown) — and since review R1 m2 a native OPC UA Acknowledge is refused up front (`BadInvalidState`) in the same window: `DriverHostActor` publishes its Primary verdict to `IDriverPrimaryRoleSource` (the native-ack gate, and the MQTT Sparkplug primary-host claim) once at start and again on every cluster membership change, not only on redundancy snapshots, so the published verdict denies as soon as a driver peer is Up rather than sitting on the single-node "allow" seed until the first snapshot. The alerts emit is skipped (the Primary peer publishes the single fleet copy). The window closes on the first delivered snapshot.
 Under warm/hot redundancy both cluster nodes run `ScriptedAlarmHostActor` and evaluate scripted alarms, keeping each node's address space and engine state warm for instant failover. To avoid duplicate rows on `/alerts` and duplicate historian writes, only the Primary node publishes externally:
 - **`alerts` topic emission** — `ScriptedAlarmHostActor` and `DriverHostActor.ForwardNativeAlarm` subscribe to the `redundancy-state` DPS topic and cache the local node's `RedundancyRole`, then gate the cluster `alerts` publish through `PrimaryGatePolicy` (table above). The OPC UA condition-node write and inbound ack/shelve command processing remain **ungated** on both nodes so the secondary is always ready to serve clients after a failover.
 - **`HistorianAdapterActor` historization** — likewise Primary-gated so alarm historization is exactly-once across all alarm sources. The actor subscribes to the `alerts` DPS topic and translates each `AlarmTransitionEvent` → `AlarmHistorianEvent` before enqueuing it on the sink; scripted alarms therefore historize exactly once regardless of cluster size.
- **The alarm sink's DRAIN** — a second, independent gate on the same decision. The enqueue gate above is not sufficient any more: since Phase 2 the store-and-forward buffer lives in the replicated LocalDb, so the Secondary holds a full copy of the Primary's queued rows and an ungated drain there would re-deliver every event continuously. `LocalDbStoreAndForwardSink` takes a `drainGate` fed from `IRedundancyRoleView` — a singleton `DriverHostActor` publishes its `PrimaryGatePolicy` verdict to, because the drain runs on a timer and cannot receive `RedundancyStateChanged` itself. A gated-off node reports `HistorianDrainState.NotPrimary`. Delivery is **at-least-once across a failover** by design: rows the old Primary delivered just before dying may not have replicated their deletes yet. See [AlarmHistorian.md](AlarmHistorian.md).
+- **The alarm sink's DRAIN** — a second, independent gate on the same decision. The enqueue gate above is not sufficient any more: since Phase 2 the store-and-forward buffer lives in the replicated LocalDb, so the Secondary holds a full copy of the Primary's queued rows and an ungated drain there would re-deliver every event continuously. `LocalDbStoreAndForwardSink` takes a `drainGate` fed from `IRedundancyRoleView` — a singleton `DriverHostActor` publishes its `PrimaryGatePolicy` verdict to, because the drain runs on a timer and cannot receive `RedundancyStateChanged` itself. A gated-off node reports `HistorianDrainState.NotPrimary`. The gate applies only to a node whose queue a peer actually holds — `LocalDbReplicationPolicy.SharesRowsWithPeer`: `LocalDb:Replication:ApiKey` AND (`PeerAddress` OR `LocalDb:SyncListenPort` > 0); any other node always drains (review R1 m3). Delivery is **at-least-once across a failover** by design: rows the old Primary delivered just before dying may not have replicated their deletes yet. See [AlarmHistorian.md](AlarmHistorian.md).
 - **Continuous-historization value CAPTURE** (Gitea #537) — a third gate on the same role rule, and the
  one that gates the *opposite* half of its pipeline. Both pair nodes serve the same address space and see
@@ -622,7 +622,10 @@ and [AlarmHistorian.md](AlarmHistorian.md). Phase 1 caches the **deployed-config
 artifact** (chunked). Its job is a single failure mode redundancy does not otherwise cover: a driver
 node restarting into a **central-SQL-Server outage** would come up with no configuration at all. With
 the cache, it **boots from the last artifact it applied** instead, and logs a running-from-cache
-signal.
+signal. A cached artifact whose bytes are empty or malformed is **no answer** (#601): nothing is applied,
 the node does **not** adopt that revision or claim running-from-cache, and it goes **Stale** (retrying the
 ConfigDb) — so a later dispatch of that same revision really applies instead of being acked as already
 current (review R1 m4). A `FetchAndCache` node lands Steady-with-no-revision instead, as for an empty cache.
 - **What replicates.** The cache can *optionally* replicate to the node's redundant pair peer over
  the LocalDb library's gRPC sync (HLC-stamped, last-writer-wins). Two tables replicate:
@@ -118,7 +118,8 @@ rationale is recorded in the engine's class doc so it is not re-proposed:
 - `ApplyOneShotShelve` / `ApplyTimedShelve(unshelveAtUtc)` / `ApplyUnshelve` — shelving transitions. `Timed` requires `unshelveAtUtc > nowUtc`.
 - `ApplyEnable` / `ApplyDisable` — operator enable/disable. Disabled alarms ignore predicate results until re-enabled; on enable, `ActiveState` is re-derived from the next evaluation.
 - `ApplyAddComment(text)` — append-only audit entry, no state change.
- `ApplyShelvingCheck(nowUtc)` — called by the 5s timer; promotes expired `Timed` shelving to `Unshelved` with a `system / AutoUnshelve` audit entry.
+- `ApplyShelvingCheck(nowUtc)` — called by the 5s timer; promotes expired `Timed` shelving to `Unshelved` with a `system / AutoUnshelve` audit entry. The sole owner of timed-shelve expiry — the OPC UA node's own SDK timer does nothing.
 - `ApplyAcknowledge` on an already-acknowledged alarm is a no-op **unless it carries a comment**, which is recorded as an `AddComment` (emission `CommentAdded`; `LastAck*` untouched) — the OPC UA node reports a commented re-ack, so the system of record must hold the comment too.
 Two invariants the machine enforces:
@@ -354,11 +355,11 @@ On allow, the handler publishes a `Commons.OpcUa.AlarmCommand` (containing comma
 - **Scripted conditions** — all eight Part 9 operations (including Enable/Disable) route through `AlarmCommandRouter` onto the `alarm-commands` topic, which `ScriptedAlarmHostActor` dispatches to the engine (`EnableAsync` / `DisableAsync` / `AcknowledgeAsync` / …). On enable, `ActiveState` is re-derived from the next predicate evaluation.
 - **Native (driver-fed) conditions** — `OnAcknowledge` branches to `NativeAlarmAckRouter` and routes the ack to the owning driver rather than the scripted engine (see §[Native-alarm OPC UA operator operations](#native-alarm-opc-ua-operator-operations) above). `OnEnableDisable`, `OnConfirm` and `OnShelve` return `BadNotSupported` immediately — a native condition supports Acknowledge and AddComment only (no engine-side surface, no upstream path, and the driver projection owns its Confirmed/Shelving; Gitea #587 review).
-`OnTimedUnshelve` (the SDK's internal auto-unshelve timer) bypasses the client gate — it is system-initiated and not subject to operator role checks.
+`OnTimedUnshelve` (the SDK's internal auto-unshelve timer, armed for the real remaining time since Gitea #597) is a **no-op**: the engine owns timed-shelve expiry via `ApplyShelvingCheck` (`system / AutoUnshelve`), and routing an `Unshelve` from the timer pre-empted that on every expiry as an operator-style unshelve from both pair nodes (review R1 M2).
 ### Delta-gate de-duplication
-`WriteAlarmCondition` fires a Part 9 condition event only when the incoming state differs from the node's current live state. For an OPC UA method call (Acknowledge, Confirm, AddComment, the shelve methods, Enable/Disable) the SDK applies the new state to the node, and the node manager then reports **the one** condition event for it from the method path (Gitea #587 — the SDK reports none of its own, since `AutoReportStateChanges` stays false; the shelve methods' state is applied by the node manager, because the SDK's never applies it; a call that changed nothing — e.g. an uncommented re-ack — reports nothing; a re-ack carrying a comment reports that comment once, Gitea #597). Native (driver-fed) conditions support only Acknowledge and AddComment; their Confirm, shelve and Enable/Disable methods return `BadNotSupported`, and their Acknowledge is refused with `BadInvalidState` on a node that is not the redundancy Primary (Gitea #597). A timed shelve's `UnshelveTime` counts down to the engine's expiry (the projection carries `UnshelveAtUtc`). The engine's later re-projection of the same transition finds the node already in that state and is suppressed, so each accepted call yields exactly one event. See `docs/AlarmTracking.md` §"One condition event per method call".
+`WriteAlarmCondition` fires a Part 9 condition event only when the incoming state differs from the node's current live state. For an OPC UA method call (Acknowledge, Confirm, AddComment, the shelve methods, Enable/Disable) the SDK applies the new state to the node, and the node manager then reports **the one** condition event for it from the method path (Gitea #587 — the SDK reports none of its own, since `AutoReportStateChanges` stays false; the shelve methods' state is applied by the node manager, because the SDK's never applies it; a call that changed nothing — e.g. an uncommented re-ack — reports nothing; a re-ack carrying a comment reports that comment once, Gitea #597, and the engine records it as an `AddComment` so it is persisted and replicated, review R1 m1). Native (driver-fed) conditions support only Acknowledge and AddComment; their Confirm, shelve and Enable/Disable methods return `BadNotSupported`, and their Acknowledge is refused with `BadInvalidState` on a node that is not the redundancy Primary (Gitea #597). A timed shelve's `UnshelveTime` counts down to the engine's expiry (the projection carries `UnshelveAtUtc`). The engine's later re-projection of the same transition finds the node already in that state and is suppressed, so each accepted call yields exactly one event. See `docs/AlarmTracking.md` §"One condition event per method call".
 ### AdminUI path
@@ -255,6 +255,11 @@ driver node**:
  reconnect loses whatever transitions happened while disconnected, same as a DPS resubscribe would.
  The cache is evicted per instance (`ITelemetryLocalHub.Evict`) when `DriverHostActor.StopChild` removes a
  driver from the plan (#609), so a decommissioned instance's last state is not replayed to new subscribers.
  `StopChild` first releases the instance's resilience state (`IDriverCapabilityInvokerFactory.Release` →
  pipeline-builder + `DriverResilienceStatusTracker.Invalidate`): `DriverResilienceStatusPublisherService`
  re-emits the whole tracker every 5 s, so evicting only the hub was undone within one tick (review R1 M1).
  Residual: a Health emit already in flight from the stopping child can land after the eviction and
  re-cache that one snapshot — a narrow race, not a recurring replay.
 - **Connection indicator** — the `/alerts` and `/script-log` live/disconnected pill is driven by
  **aggregate** stream health across all dialers: connected when **at least one** node stream is
  up, disconnected when all are down. This matches today's DPS `SubscribeAck`/`PostStop` pill
@@ -92,7 +92,9 @@ Replication has two knobs, both under `LocalDb`:
 > 2026-08-18 `LocalDbSetup.OnReady` installs the change-capture triggers only when the node is
 > configured to replicate — the predicate is **`PeerAddress` OR `ApiKey` non-blank** (the OR because
 > the passive node has no `PeerAddress`; `SyncListenPort` is deliberately not part of it, since a
-> port with no key is fail-closed and syncs with nobody). Consequences at enablement time:
+> port with no key is fail-closed and syncs with nobody). The alarm-historian drain gate uses the
 > stricter `ApiKey` AND (`PeerAddress` OR `SyncListenPort > 0`) — so set the key **and** the port/peer
 > together, or the node keeps draining its own queue. Consequences at enablement time:
 >
 > - A configured node registers with **`baselineExistingRows: true`**, so the rows **already in the
 >   file** are seeded into the row-version ledger at the LWW floor (HLC `0`) and actually travel.
@@ -76,7 +76,7 @@ public sealed class ClusterRoleInfo : IClusterRoleInfo, IDisposable
            }
        }
-        SeedFromCurrentState();
+        SeedFrom(_cluster.State);
        _subscriber = system.ActorOf(Props.Create(() => new SubscriberActor(this)), "clusterroleinfo-subscriber");
    }
@@ -121,12 +121,19 @@ public sealed class ClusterRoleInfo : IClusterRoleInfo, IDisposable
    /// <summary>Occurs when the leader for a role changes.</summary>
    public event EventHandler<RoleLeaderChangedEventArgs>? RoleLeaderChanged;
-    private void SeedFromCurrentState()
+    /// <summary>Seeds the member/leader maps from a cluster-state snapshot (the constructor passes the live
    /// <c>Cluster.State</c>; exposed for tests).</summary>
    /// <param name="snapshot">The cluster-state snapshot.</param>
    internal void SeedFrom(ClusterEvent.CurrentClusterState snapshot)
    {
        var snapshot = _cluster.State;
        lock (_lock)
        {
-            foreach (var member in snapshot.Members)
+            // Cluster-002 (review R1 m6): seed only USABLE members. A member already Leaving / Exiting / Down
            // (or still Joining) at seed time is not one, and a departing member never fires a later
            // MemberLeft / MemberExited / MemberDowned that HandleMemberEvent would drop it on — it would
            // linger until MemberRemoved, the very delay the member-event handling exists to avoid. A
            // Joining member is added by its MemberUp.
            foreach (var member in snapshot.Members.Where(m => m.Status is MemberStatus.Up or MemberStatus.WeaklyUp))
                foreach (var role in member.Roles)
                {
                    if (!_membersByRole.TryGetValue(role, out var set))
@@ -28,6 +28,16 @@ public interface IDriverCapabilityInvokerFactory
    /// </param>
    /// <returns>A per-instance <see cref="IDriverCapabilityInvoker"/>.</returns>
    IDriverCapabilityInvoker Create(string driverInstanceId, string driverType, string? resilienceConfigJson = null);
    /// <summary>
    ///     Release every piece of shared resilience state held for <paramref name="driverInstanceId"/> — called
    ///     by the host when the instance leaves the deployed plan (#609). A concrete factory drops the
    ///     instance's cached pipelines and its status-tracker snapshots, so the periodic resilience publisher
    ///     stops re-emitting (and re-priming new telemetry subscribers with) a decommissioned driver's last
    ///     breaker/failure counters. Default: no-op (a factory that holds no state).
    /// </summary>
    /// <param name="driverInstanceId">The driver instance that was removed.</param>
    void Release(string driverInstanceId) { }
 }
 /// <summary>
@@ -105,7 +105,17 @@ public static class Part9StateMachine
            throw new ArgumentException("User identity required for audit.", nameof(user));
        if (current.Acked == AlarmAckedState.Acknowledged)
-            return TransitionResult.NoOp(current, "already acknowledged");
+        {
            // A COMMENTED re-ack carries a real comment even though nothing is acknowledged (Gitea #597 review
            // m1): the OPC UA node reports it — the SDK writes the comment on a re-ack — so the system of
            // record must hold it too, or the event advertises a comment that exists only on the node that
            // served the call (never persisted, never replicated to the pair peer). Recorded exactly like
            // AddComment; the ack fields (LastAck*) stay those of the real acknowledge. An uncommented re-ack
            // is still a no-op.
            return string.IsNullOrWhiteSpace(comment)
                ? TransitionResult.NoOp(current, "already acknowledged")
                : ApplyAddComment(current, user, comment, nowUtc);
        }
        var audit = AppendComment(current.Comments, nowUtc, user, "Acknowledge", comment);
        var next = current with
@@ -90,4 +90,16 @@ public sealed class DriverCapabilityInvokerFactory : IDriverCapabilityInvokerFac
            statusTracker: _statusTracker,
            optionsGeneration: generation);
    }
    /// <inheritdoc />
    /// <remarks>
    ///     #609: without this the tracker kept a removed instance's snapshots forever (only <see cref="Create"/>
    ///     invalidated, and a removed instance is never re-created), and the resilience publisher re-emitted
    ///     them every tick — undoing the host's telemetry-hub eviction within one interval.
    /// </remarks>
    public void Release(string driverInstanceId)
    {
        _builder.Invalidate(driverInstanceId);
        _statusTracker.Invalidate(driverInstanceId);
    }
 }
@@ -46,7 +46,7 @@ public static class LocalDbRegistration
    ///     Configuration key carrying the dedicated h2c sync listener's port. Zero or absent means
    ///     no listener is bound and the passive sync endpoint is not mapped.
    /// </summary>
-    public const string SyncListenPortKey = "LocalDb:SyncListenPort";
+    public const string SyncListenPortKey = Runtime.LocalDbReplicationPolicy.SyncListenPortKey;
    /// <summary>
    ///     The port the dedicated cleartext-HTTP/2 sync listener should bind, or <c>0</c> when the
@@ -170,9 +170,9 @@ public static class LocalDbSetup
    /// <returns><c>true</c> when this node is configured to replicate.</returns>
    public static bool ReplicationIsConfigured(IConfiguration configuration)
    {
-        // Shared with the alarm-historian drain gate (#603): one predicate, so the nodes that capture
+        // Defined beside the alarm-historian drain gate's stricter predicate (#603, review R1 m3): every node
-        // changes and the nodes that stand down for a peer can never disagree.
+        // the gate treats as sharing rows also captures, so the two never disagree in the harmful direction.
-        return Runtime.LocalDbReplicationPolicy.IsConfigured(configuration);
+        return Runtime.LocalDbReplicationPolicy.CapturesChanges(configuration);
    }
    /// <summary>Opts every replicated table into change capture, baselining what is already there.</summary>
@@ -749,7 +749,10 @@ if (LogPipelineDiagnostics.CheckFileSinkGrowth(app.Configuration,
 await app.RunAsync();
 }
-catch (Exception ex)
+// HostAbortedException is how design-time tooling (EF `dotnet ef … --startup-project`, OpenAPI generators —
 // HostFactoryResolver) stops Program once it has the built host. It is not a failure: letting it reach the
 // Fatal + Environment.Exit(1) below would kill the tool's own process (#618, review R1 m5).
 catch (Exception ex) when (ex is not HostAbortedException)
 {
    // Log.Logger is still the silent bootstrap logger when the failure happened before AddZbSerilog
    // ran (e.g. EffectiveRoles.Resolve), so ALSO write to stderr — the one channel that always exists.
@@ -874,8 +874,7 @@ public sealed class OtOpcUaNodeManager : CustomNodeManager2
                    // (shelved, oneShot) flag pair. Timed shelving's expiry is owned by the engine; the time
                    // passed here is the engine's remaining time, so the node's UnshelveTime counts down to the
                    // real expiry (Gitea #597 — a 0 here reads as MaxTimeShelved, ~24.8 days). The SDK timer it
-                    // arms only routes an Unshelve back to the engine (OnTimedUnshelve), a no-op once the
+                    // arms does nothing (OnTimedUnshelve is a no-op): the engine's own expiry re-projects.
                    // engine's own expiry has run.
                    condition.SetShelvingState(
                        SystemContext,
                        shelved: state.Shelving != AlarmShelvingKind.Unshelved,
@@ -1375,25 +1374,17 @@ public sealed class OtOpcUaNodeManager : CustomNodeManager2
                    : ("TimedShelve", DateTime.UtcNow + TimeSpan.FromMilliseconds(shelvingTime));
                return HandleAlarmCommand(context, condition, operation, comment: null, unshelveAt);
            };
-            // The auto-unshelve timer callback is SDK-initiated (the TimedShelve duration expired); the SDK
+            // The SDK's auto-unshelve timer (armed by SetShelvingState for the real remaining time since Gitea
-            // fires it with the node manager's system context — there is NO session and NO user identity.
+            // #597) fires with a session-less system context. It is deliberately a NO-OP: the engine owns
-            // Routing through HandleAlarmCommand would hit the AlarmAck gate and return BadUserAccessDenied,
+            // timed-shelve expiry (ScriptedAlarmEngine's shelving check → Part9StateMachine.ApplyShelvingCheck,
-            // leaving the alarm permanently shelved. Instead, bypass the client gate, extract the AlarmId the
+            // recorded as "AutoUnshelve" by "system" with its "Timed shelving expired" comment), and its
-            // same way HandleAlarmCommand does, and route an Unshelve command so the engine clears its shelve
+            // re-projection is what unshelves this node. Routing an operator-style "Unshelve" from here used to
-            // state. The manual-client Unshelve path goes through OnShelve(shelving:false) and stays gated.
+            // pre-empt that check on every expiry — from BOTH pair nodes, rewriting the audit trail as a manual
-            alarm.OnTimedUnshelve = (context, condition) =>
+            // unshelve — and a timer callback already queued when a re-shelve disposed it could unshelve the
-            {
+            // fresh shelve. The SDK's own OnTimerExpired changes no shelving state itself (it only refreshes
-                var alarmId = condition.NodeId.Identifier?.ToString() ?? string.Empty;
+            // UnshelveTime after this callback), so returning without routing leaves the node shelved until
-                // User MUST be non-empty: the engine's Part9StateMachine.ApplyUnshelve rejects a
+            // the engine's expiry lands. Native conditions are never timed-shelved (BadNotSupported above).
-                // null/whitespace user (ArgumentException "User required."), and that exception is swallowed
+            alarm.OnTimedUnshelve = (_, _) => ServiceResult.Good;
                // downstream — an empty user would make the timed auto-unshelve silently no-op, so a
                // TimedShelve would never auto-expire. There is no client principal on a system-timer
                // unshelve, so label it the canonical engine-internal "system" user (matching the engine's
                // own AutoUnshelve audit user and the codebase-wide "system" system-actor convention). The
                // AlarmAck gate bypass is intentional and preserved — this is a session-less SDK timer.
                AlarmCommandRouter?.Invoke(new AlarmCommand(alarmId, "Unshelve", "system", null, null));
                return ServiceResult.Good;
            };
            // H4 — inbound Part 9 Enable/Disable (the condition type's built-in Enable/Disable methods route
            // here via this delegate). The engine handles Enable/Disable for SCRIPTED alarms
@@ -1596,9 +1587,11 @@ public sealed class OtOpcUaNodeManager : CustomNodeManager2
    ///         <b>Only a real change is reported.</b> The condition's gate-relevant state is read before and after
    ///         the SDK handler, and nothing is reported when it did not move. The SDK accepts acknowledging an
    ///         acknowledged condition and sets <c>Confirmed = false</c> again; a re-ack therefore also restores the
-    ///         pre-call Confirmed and Retain, so a scripted node never drifts from its engine (which ignores a
+    ///         pre-call Confirmed and Retain, so a scripted node never drifts from its engine (which ignores an
-    ///         re-ack) and the call is silent — unless it carried a comment, which the SDK writes and which is then
+    ///         uncommented re-ack) and the call is silent — unless it carried a comment, which the SDK writes and
-    ///         reported once like AddComment (Gitea #597). AddComment always reports: adding a comment is itself
+    ///         which is then reported once like AddComment (Gitea #597). The scripted engine records that comment
    ///         too (<c>Part9StateMachine.ApplyAcknowledge</c> treats a commented re-ack as AddComment), so the
    ///         reported comment is persisted and replicated, not node-local. AddComment always reports: adding a comment is itself
    ///         the change. A NATIVE re-ack is still routed to the driver: the route is fire-and-forget, so a node
    ///         showing Acked does not prove the upstream system acknowledged, and re-acking is the documented retry.
    ///     </para>
@@ -1635,7 +1628,7 @@ public sealed class OtOpcUaNodeManager : CustomNodeManager2
                // The requested duration, so UnshelveTime counts down to the real expiry (Gitea #597 — 0 reads as
                // MaxTimeShelved, ~24.8 days). The scripted engine still owns the expiry and re-projects Unshelved
                // when it passes (with its own remaining time meanwhile — WriteAlarmCondition); the SDK timer this
-                // arms only routes an Unshelve back to it (OnTimedUnshelve), a no-op once the engine has expired it.
+                // arms does nothing (OnTimedUnshelve is a no-op — the engine's AutoUnshelve owns the expiry).
                timedShelve.OnCall = (context, method, objectId, shelvingTime) => ReportAfterMethod(
                    () => timed(context, method, objectId, shelvingTime),
                    () => alarm.SetShelvingState(SystemContext, shelved: true, oneShot: false, shelvingTime: shelvingTime));
@@ -788,9 +788,91 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        // Same rationale for the ScriptedAlarm host — the bootstrap-restore path Tells it
        // ApplyScriptedAlarms through the same PushDesiredSubscriptions pass.
        SpawnScriptedAlarmHost();
        StartPrimaryVerdictPublishing();
        Bootstrap();
    }
    /// <summary>
    ///     Review R1 m2 (Gitea #597): the published Primary data-plane verdict (<see cref="IDriverPrimaryRoleSource"/>
    ///     — the native-ack gate and the Sparkplug primary-host claim) must mirror <see cref="ShouldServiceAsPrimary"/>
    ///     from the moment this host starts, not only from the first redundancy snapshot. Before this, the view sat
    ///     on its single-node seed (allow) through the whole boot window, so on a pair BOTH nodes accepted a native
    ///     ack while this host — which evaluates the member count live — dropped the upstream route on the
    ///     non-Primary: the acked → unacked revert #597 targets. So publish once now, and again whenever cluster
    ///     membership changes (a driver peer coming Up turns an unknown role from allow into deny before any
    ///     snapshot names this node). A non-cluster ActorRefProvider (legacy/test harnesses) just keeps the
    ///     initial publish.
    /// </summary>
    private void StartPrimaryVerdictPublishing()
    {
        if (_primaryRoleSource is null) return;
        PublishPrimaryVerdict();
        try
        {
            Akka.Cluster.Cluster.Get(Context.System).Subscribe(
                Self, Akka.Cluster.ClusterEvent.SubscriptionInitialStateMode.InitialStateAsSnapshot,
                typeof(Akka.Cluster.ClusterEvent.IMemberEvent));
            _subscribedToMemberEvents = true;
        }
        catch (Exception)
        {
            // Non-cluster ActorRefProvider — no membership to follow; DriverMemberCount() reads 0 there too.
        }
    }
    /// <summary>A cluster member changed status: re-publish the Primary verdict now and once more shortly after.
    /// The deferred pass covers the cluster read-view (what <see cref="DriverMemberCount"/> reads) updating on its
    /// own subscriber AFTER this event was delivered here — the two are not ordered.</summary>
    private void OnDriverMembershipChanged(Akka.Cluster.ClusterEvent.IMemberEvent _)
    {
        PublishPrimaryVerdictIfChanged();
        // One keyed timer: a burst of member events collapses into a single deferred pass.
        Timers.StartSingleTimer("republish-primary-verdict", RepublishPrimaryVerdict.Instance, MembershipRepublishDelay);
    }
    /// <summary>Publishes <see cref="ShouldServiceAsPrimary"/> to the primary-role source (no-op when none is wired).
    /// The snapshot path publishes on every snapshot, as the other verdicts do.</summary>
    private void PublishPrimaryVerdict()
    {
        if (_primaryRoleSource is null) return;
        var verdict = ShouldServiceAsPrimary();
        _lastPublishedPrimaryVerdict = verdict;
        _primaryRoleSource.PublishPrimaryRole(verdict);
    }
    /// <summary>The membership path publishes only a CHANGED verdict — member events are frequent and carry no
    /// role information, so republishing an unchanged verdict on each would be noise.</summary>
    private void PublishPrimaryVerdictIfChanged()
    {
        if (_primaryRoleSource is null || ShouldServiceAsPrimary() == _lastPublishedPrimaryVerdict) return;
        PublishPrimaryVerdict();
    }
    /// <inheritdoc />
    protected override void PostStop()
    {
        if (_subscribedToMemberEvents)
        {
            try { Akka.Cluster.Cluster.Get(Context.System).Unsubscribe(Self); }
            catch (Exception) { /* system shutting down */ }
        }
        base.PostStop();
    }
    /// <summary>Self-message: re-evaluate and re-publish the Primary verdict (see <see cref="OnDriverMembershipChanged"/>).</summary>
    internal sealed class RepublishPrimaryVerdict
    {
        /// <summary>The shared instance.</summary>
        public static readonly RepublishPrimaryVerdict Instance = new();
        private RepublishPrimaryVerdict() { }
    }
    /// <summary>How long after a membership event the Primary verdict is re-published a second time.</summary>
    private static readonly TimeSpan MembershipRepublishDelay = TimeSpan.FromSeconds(1);
    private bool _subscribedToMemberEvents;
    private bool? _lastPublishedPrimaryVerdict;
    /// <summary>
    ///     Spawns the single <see cref="VirtualTagHostActor"/> child that owns the Equipment-namespace
    ///     VirtualTagActors and bridges their results onto the OPC UA publish actor. A test-supplied
@@ -990,12 +1072,24 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
            var deploymentId = DeploymentId.Parse(cached.DeploymentId);
            var revision = RevisionHash.Parse(cached.RevisionHash);
            _currentRevision = revision;
            BecomeSteady();
            // Restore drivers + address space + subscriptions from the cached bytes — no SQL read, no
            // re-ack (the deployment is already Applied). Same in-hand-bytes apply as the Direct-mode
            // cache boot, but this is the steady-state config source, not an outage fallback.
-            ApplyCachedArtifact(deploymentId, cached.Artifact);
+            if (!ApplyCachedArtifact(deploymentId, cached.Artifact))
            {
                // #601 / review R1 m4: unusable cached bytes applied nothing, so they must not become the
                // current revision — that would ack a re-dispatch of the same revision as already applied
                // without ever applying it. Steady-with-no-revision, exactly as for an empty cache.
                _log.Warning(
                    "DriverHost {Node}: FetchAndCache boot — the cached artifact for deployment {Id} (rev {Rev}) " +
                    "is unreadable; entering Steady (no revision). The next dispatch will fetch.",
                    _localNode, deploymentId, revision);
                BecomeSteady();
                return;
            }
            _currentRevision = revision;
            BecomeSteady();
            _log.Info(
                "DriverHost {Node}: FetchAndCache boot — restored served state for deployment {Id} " +
@@ -1059,14 +1153,24 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
            var deploymentId = DeploymentId.Parse(cached.DeploymentId);
            var revision = RevisionHash.Parse(cached.RevisionHash);
            // #601 / review R1 m4: unusable cached bytes applied nothing — the node is NOT serving a
            // configuration, so it must not claim RUNNING FROM CACHE, adopt the revision (a re-dispatch of
            // it would then be acked without being applied) or skip Stale's ConfigDb retry loop.
            if (!ApplyCachedArtifact(deploymentId, cached.Artifact))
            {
                _log.Error(
                    "DriverHost {Node}: the cached artifact for deployment {Id} (rev {Rev}) is unreadable; " +
                    "not booting from it — entering Stale with no configuration.",
                    _localNode, deploymentId, revision);
                return false;
            }
            // Steady, not Stale: this node is serving a real configuration. The retry-db timer that
            // Stale would have started does not run here, so recovery rides on the next dispatch —
            // matching how a normally-booted node behaves.
            _currentRevision = revision;
            BecomeSteady();
            ApplyCachedArtifact(deploymentId, cached.Artifact);
            _isRunningFromCache = true;
            _log.Warning(
                "DriverHost {Node}: RUNNING FROM CACHE — central ConfigDb is unreachable, so this node " +
@@ -1094,16 +1198,16 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
    ///     <c>UpsertNodeDeploymentState</c> and <c>SendAck</c> for the same reason: both write to or
    ///     depend on central.
    /// </remarks>
-    private void ApplyCachedArtifact(DeploymentId deploymentId, byte[] blob)
+    /// <returns><see langword="false"/> when the bytes were empty or malformed and nothing was applied — the
    /// caller must not adopt the cached revision (#601, review R1 m4).</returns>
    private bool ApplyCachedArtifact(DeploymentId deploymentId, byte[] blob)
    {
        var correlation = CorrelationId.NewId();
        // #601: null = the cached bytes were empty or malformed — no answer, not an empty configuration.
        // Keep last-known-good: no address-space rebuild and no subscription push from unusable bytes.
        // #601: null = the cached bytes were empty or malformed — no answer, not an empty configuration.
        // Keep last-known-good: no address-space rebuild and no subscription push from unusable bytes.
        if (ReconcileDriversFromBlob(deploymentId, blob) is null)
-            return;
+            return false;
        // Hand the cached blob to the rebuild so it materialises the client-facing address space from
        // it directly. Passing only the deploymentId would drive a ConfigDb read — unreachable here by
@@ -1113,6 +1217,7 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
            new ZB.MOM.WW.OtOpcUa.Runtime.OpcUa.OpcUaPublishActor.RebuildAddressSpace(correlation, deploymentId, blob));
        PushDesiredSubscriptionsFromArtifact(deploymentId, blob);
        return true;
    }
    /// <summary>
@@ -1170,6 +1275,9 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        Receive<RouteNodeWrite>(HandleRouteNodeWrite);
        Receive<RouteNativeAlarmAck>(HandleRouteNativeAlarmAck);
        Receive<RedundancyStateChanged>(OnRedundancyStateChanged);
        Receive<Akka.Cluster.ClusterEvent.IMemberEvent>(OnDriverMembershipChanged);
        Receive<RepublishPrimaryVerdict>(_ => PublishPrimaryVerdictIfChanged());
        Receive<Akka.Cluster.ClusterEvent.CurrentClusterState>(_ => PublishPrimaryVerdictIfChanged());
        // The redundancy-state topic also carries OpcUaProbeResult (OpcUaPublishActor peer-probes).
        // We don't consume it here — drop it so it doesn't dead-letter (matches PeerProbeSupervisor).
        Receive<ZB.MOM.WW.OtOpcUa.Runtime.Health.PeerOpcUaProbeActor.OpcUaProbeResult>(_ => { });
@@ -1228,6 +1336,9 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        Receive<RouteNodeWrite>(HandleRouteNodeWrite);
        Receive<RouteNativeAlarmAck>(HandleRouteNativeAlarmAck);
        Receive<RedundancyStateChanged>(OnRedundancyStateChanged);
        Receive<Akka.Cluster.ClusterEvent.IMemberEvent>(OnDriverMembershipChanged);
        Receive<RepublishPrimaryVerdict>(_ => PublishPrimaryVerdictIfChanged());
        Receive<Akka.Cluster.ClusterEvent.CurrentClusterState>(_ => PublishPrimaryVerdictIfChanged());
        // The redundancy-state topic also carries OpcUaProbeResult (OpcUaPublishActor peer-probes).
        // We don't consume it here — drop it so it doesn't dead-letter (matches PeerProbeSupervisor).
        Receive<ZB.MOM.WW.OtOpcUa.Runtime.Health.PeerOpcUaProbeActor.OpcUaProbeResult>(_ => { });
@@ -1975,7 +2086,7 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        // drain verdict published immediately above. A driver reading this claims an exclusive identity
        // on a shared external system (Gitea #511: the Sparkplug host id), where two claimants is the
        // harm, so an unknown role with a driver peer present must deny.
-        _primaryRoleSource?.PublishPrimaryRole(ShouldServiceAsPrimary());
+        PublishPrimaryVerdict();
    }
    private void Stale()
@@ -1998,6 +2109,9 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        Receive<ReprovisionDeferred>(m => ReprovisionChild(m.Request, m.Child));
        Receive<ReconnectDriver>(HandleReconnectDriver);
        Receive<RedundancyStateChanged>(OnRedundancyStateChanged);
        Receive<Akka.Cluster.ClusterEvent.IMemberEvent>(OnDriverMembershipChanged);
        Receive<RepublishPrimaryVerdict>(_ => PublishPrimaryVerdictIfChanged());
        Receive<Akka.Cluster.ClusterEvent.CurrentClusterState>(_ => PublishPrimaryVerdictIfChanged());
        // The redundancy-state topic also carries OpcUaProbeResult (OpcUaPublishActor peer-probes).
        // We don't consume it here — drop it so it doesn't dead-letter (matches PeerProbeSupervisor).
        Receive<ZB.MOM.WW.OtOpcUa.Runtime.Health.PeerOpcUaProbeActor.OpcUaProbeResult>(_ => { });
@@ -3233,7 +3347,10 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
        Context.Stop(entry.Actor);
        _children.Remove(driverInstanceId);
        // #609: the instance left the plan (StopChild is only reached via plan.ToStop), so its cached
-        // Health/Resilience snapshots must not replay to new telemetry subscribers.
+        // Health/Resilience snapshots must not replay to new telemetry subscribers. Release the shared
        // resilience state FIRST: the resilience publisher re-emits the tracker's snapshots every tick, so
        // evicting the hub alone was undone within one interval.
        _invokerFactory.Release(driverInstanceId);
        _telemetryHub?.Evict(driverInstanceId);
        _log.Info("DriverHost {Node}: stopped driver child {Id}", _localNode, driverInstanceId);
    }
@@ -3,26 +3,36 @@ using Microsoft.Extensions.Configuration;
 namespace ZB.MOM.WW.OtOpcUa.Runtime;
 /// <summary>
-///     The ONE definition of "this node's LocalDb is replicated to a peer" (#603). Two consumers must agree
+///     The definitions of "this node's LocalDb is replicated to a peer" (#603) — one place, two deliberately
-///     on it, and used to carry separate predicates that diverged: the Host's table-registration step
+///     different predicates for the two consumers: the Host's table-registration step
-///     (<c>LocalDbSetup.ReplicationIsConfigured</c> - which nodes capture changes) and the alarm-historian
+///     (<c>LocalDbSetup.ReplicationIsConfigured</c> → <see cref="CapturesChanges"/>: which nodes capture
-///     drain gate (<c>AddAlarmHistorian</c> - which nodes stand down in favour of a peer). Lives in Runtime
+///     changes) and the alarm-historian drain gate (<c>AddAlarmHistorian</c> → <see cref="SharesRowsWithPeer"/>:
-///     because Host references Runtime and not the reverse.
+///     which nodes stand down in favour of a peer). Lives in Runtime because Host references Runtime and not
 ///     the reverse.
 /// </summary>
 /// <remarks>
 ///     <para>
-///         <b>Predicate: <c>LocalDb:Replication:PeerAddress</c> OR <c>LocalDb:Replication:ApiKey</c> is
+///         <b>Why two predicates (review R1 m3).</b> The two answers cost differently when wrong. Registering a
-///         non-blank.</b> The OR is required: only the dialling half of a pair sets <c>PeerAddress</c>, while
+///         node that has no working sync path only costs capture triggers. Gating the drain of a node that
-///         the passive half carries the shared <c>ApiKey</c> (plus a listen port) and nothing else.
+///         shares its queue with nobody parks its alarm events behind a Primary that does not hold them —
 ///         undelivered until the node is promoted, and drop-oldest at <c>Capacity</c> meanwhile. So
 ///         registration stays permissive and the gate is strict. Every node the gate treats as sharing also
 ///         captures, so the two never disagree in the harmful direction.
 ///     </para>
 ///     <para>
-///         <b>Why the drain gate takes the same predicate (it previously used PeerAddress OR
+///         <b>Registration: <c>LocalDb:Replication:PeerAddress</c> OR <c>LocalDb:Replication:ApiKey</c>.</b>
-///         <c>SyncListenPort</c>).</b> The gate's question is "does a peer hold the same rows?". The passive
+///         The OR is required: only the dialling half of a pair sets <c>PeerAddress</c>, while the passive
-///         sync endpoint is fail-closed on the key, so a node with a listen port but no <c>ApiKey</c> rejects
+///         half carries the shared <c>ApiKey</c> (plus a listen port) and nothing else. Keying on
-///         every inbound stream and shares its queue with nobody - gating it would park its alarm events
+///         <c>PeerAddress</c> alone would strip capture from every passive node.
-///         behind a Primary that cannot deliver them. Conversely an <c>ApiKey</c>-only passive node IS
+///     </para>
-///         registered and does receive its peer's rows, so it must be gated or both halves drain. The port
+///     <para>
-///         is not evidence of a peer; the key is.
+///         <b>Drain gate: <c>ApiKey</c> AND (<c>PeerAddress</c> OR <c>LocalDb:SyncListenPort</c> &gt; 0).</b> A
 ///         peer holds this node's rows only over a working sync path: a dial (<c>PeerAddress</c>) or a
 ///         listener (<c>SyncListenPort</c>), each authenticated by the key. Both auth directions fail closed,
 ///         so a path without the key shares nothing, and a key with neither path (e.g. a fleet-wide
 ///         secret/env <c>ApiKey</c> on a node with no peer) shares nothing either. Every docker-dev shape
 ///         gives the same answer under both predicates: site-a-1 (PeerAddress + ApiKey + port) and site-a-2
 ///         (ApiKey + port) are gated; central and site-b (neither) are not.
 ///     </para>
 /// </remarks>
 public static class LocalDbReplicationPolicy
@@ -30,14 +40,31 @@ public static class LocalDbReplicationPolicy
    /// <summary>Configuration section holding the LocalDb replication options.</summary>
    public const string ReplicationSectionPath = "LocalDb:Replication";
-    /// <summary>Whether this node's LocalDb replicates to a pair peer.</summary>
+    /// <summary>The dedicated passive-sync listener port key (same key as <c>LocalDbRegistration.SyncListenPortKey</c>).</summary>
    public const string SyncListenPortKey = "LocalDb:SyncListenPort";
    /// <summary>Whether this node registers its replicated tables for change capture (table registration).</summary>
    /// <param name="configuration">The application configuration.</param>
    /// <returns><c>true</c> when <c>PeerAddress</c> or <c>ApiKey</c> is set under the replication section.</returns>
-    public static bool IsConfigured(IConfiguration configuration)
+    public static bool CapturesChanges(IConfiguration configuration)
    {
        ArgumentNullException.ThrowIfNull(configuration);
        var section = configuration.GetSection(ReplicationSectionPath);
        return !string.IsNullOrWhiteSpace(section["PeerAddress"])
            || !string.IsNullOrWhiteSpace(section["ApiKey"]);
    }
    /// <summary>Whether a pair peer actually holds this node's replicated rows (the alarm drain gate).</summary>
    /// <param name="configuration">The application configuration.</param>
    /// <returns><c>true</c> when <c>ApiKey</c> is set AND either <c>PeerAddress</c> is set or
    ///     <c>LocalDb:SyncListenPort</c> is a positive port.</returns>
    public static bool SharesRowsWithPeer(IConfiguration configuration)
    {
        ArgumentNullException.ThrowIfNull(configuration);
        var section = configuration.GetSection(ReplicationSectionPath);
        if (string.IsNullOrWhiteSpace(section["ApiKey"])) return false;
        var dials = !string.IsNullOrWhiteSpace(section["PeerAddress"]);
        var listens = int.TryParse(configuration[SyncListenPortKey], out var port) && port > 0;
        return dials || listens;
    }
 }
@@ -177,11 +177,13 @@ public static class ServiceCollectionExtensions
        // possibly deliver its events. Scoping the gate to "is my queue actually shared?" is what
        // keeps the two scopes from disagreeing.
        // BOTH halves of a pair share the queue, but only one of them dials: the initiator sets
-        // Replication:PeerAddress, its partner carries the shared ApiKey and waits. The predicate is
+        // Replication:PeerAddress, its partner carries the shared ApiKey plus SyncListenPort and waits.
-        // LocalDbReplicationPolicy - the same one LocalDbSetup uses to decide which nodes capture
+        // The predicate is LocalDbReplicationPolicy.SharesRowsWithPeer: ApiKey AND (PeerAddress OR
-        // changes at all. It deliberately does NOT count SyncListenPort: the passive endpoint is
+        // SyncListenPort > 0) - an authenticated sync path in either direction. Both auth directions
-        // fail-closed on the key, so a port with no ApiKey shares nothing and must keep draining.
+        // fail closed, so a path without the key, or a key with no path (a fleet-wide secret on an
-        var replicated = LocalDbReplicationPolicy.IsConfigured(configuration);
+        // unpaired node), shares nothing and must keep draining. Deliberately stricter than the
        // registration predicate (CapturesChanges) - see the policy's remarks.
        var replicated = LocalDbReplicationPolicy.SharesRowsWithPeer(configuration);
        if (!replicated)
        {
@@ -97,4 +97,30 @@ public sealed class ClusterRoleInfoMemberEventTests
            return Task.CompletedTask;
        });
    }
    /// <summary>
    ///     Review R1 m6: the initial seed lists only Up / WeaklyUp members. A member already departing at seed
    ///     time never fires a later Left/Exited/Downed event, so seeding it would list it until MemberRemoved.
    ///     The Up and WeaklyUp members are the positive control (the seed does list members).
    /// </summary>
    [Fact]
    public async Task Initial_seed_lists_only_up_and_weakly_up_members()
    {
        await WithInfoAsync(info =>
        {
            var members = ImmutableSortedSet.Create<Member>(
                NewMember(21001, MemberStatus.Up),
                NewMember(21002, MemberStatus.WeaklyUp),
                NewMember(21003, MemberStatus.Leaving),
                NewMember(21004, MemberStatus.Exiting),
                NewMember(21005, MemberStatus.Down),
                NewMember(21006, MemberStatus.Joining));
            info.SeedFrom(new ClusterEvent.CurrentClusterState().Copy(members: members));
            info.MembersWithRole("driver").Select(n => n.Value).OrderBy(v => v).ShouldBe(
                ["127.0.0.1:21001", "127.0.0.1:21002"]);
            return Task.CompletedTask;
        });
    }
 }
@@ -79,6 +79,28 @@ public sealed class Part9StateMachineTests
        r.Emission.ShouldBe(EmissionKind.None);
    }
    /// <summary>
    ///     Gitea #597 review m1 — a COMMENTED re-ack records its comment (as AddComment would) so the comment the
    ///     OPC UA event advertised is durable; the original acknowledge's ack fields are untouched.
    /// </summary>
    [Fact]
    public void Commented_reack_records_the_comment_without_reacknowledging()
    {
        var active = Part9StateMachine.ApplyPredicate(Fresh(), true, T0.AddSeconds(1)).State;
        var acked = Part9StateMachine.ApplyAcknowledge(active, "alice", "first look", T0.AddSeconds(2)).State;
        var r = Part9StateMachine.ApplyAcknowledge(acked, "bob", "still on it", T0.AddSeconds(3));
        r.Emission.ShouldBe(EmissionKind.CommentAdded);
        r.State.Comments.Count.ShouldBe(2);
        r.State.Comments[1].User.ShouldBe("bob");
        r.State.Comments[1].Kind.ShouldBe("AddComment");
        r.State.Comments[1].Text.ShouldBe("still on it");
        r.State.LastAckUser.ShouldBe("alice");
        r.State.LastAckComment.ShouldBe("first look");
        r.State.LastAckUtc.ShouldBe(T0.AddSeconds(2));
    }
    /// <summary>Verifies that acknowledging without a user throws ArgumentException.</summary>
    [Fact]
    public void Acknowledge_without_user_throws()
@@ -263,14 +263,18 @@ public sealed class ConditionMethodEventTests
    /// <summary>
    ///     A scripted TimedShelve expires: the engine's expiry comes back as an Unshelved re-projection, which
-    ///     reports exactly one unshelve event after the one shelve event.
+    ///     reports exactly one unshelve event after the one shelve event — and the SDK's own expiry timer routes
    ///     nothing (review R1 M2).
    /// </summary>
    /// <remarks>
    ///     The expiry is the engine's (computed from the command's <c>UnshelveAtUtc</c>). Since Gitea #597 the node
-    ///     is shelved for the real duration, so the SDK's own timer fires at the same instant — but it only routes
+    ///     is shelved for the real duration, so the SDK's own timer fires at that instant. It used to route an
-    ///     an Unshelve back to the engine (ignored by this stand-in, a no-op on an unshelved engine alarm) and
+    ///     operator-style <c>Unshelve</c> by <c>system</c>, which — because the real engine only checks expiry every
-    ///     changes no node state, so it adds no event. The stand-in engine here schedules its expiry from the real
+    ///     5 s — pre-empted the engine's <c>AutoUnshelve</c> on every expiry. This stand-in engine therefore
-    ///     <c>UnshelveAtUtc</c> the method handler computed.
+    ///     (a) expires LATE, like the real one, so the SDK timer demonstrably fires first; and (b) ACTS on a routed
    ///     <c>Unshelve</c> as the real engine would, recording who unshelved it. The wrapped
    ///     <c>OnTimedUnshelve</c> is the positive control: the SDK timer did fire, and still nothing was routed and
    ///     the unshelve on record is the engine's own.
    /// </remarks>
    [Fact]
    public async Task A_scripted_timed_shelve_expires_with_exactly_one_unshelve_event()
@@ -283,21 +287,47 @@ public sealed class ConditionMethodEventTests
            rig.MaterialiseScripted();
            var current = PreState("TimedShelve", native: false) with { Message = "base" };
            Task? expiry = null;
            var routed = new System.Collections.Concurrent.ConcurrentQueue<string>();
            string? unshelvedBy = null;
            var engineLag = TimeSpan.FromMilliseconds(1500); // the real engine's 5 s shelving-check granularity
            rig.Nm.AlarmCommandRouter = cmd =>
            {
                routed.Enqueue(cmd.Operation);
                if (cmd.Operation == "Unshelve")
                {
                    // What the real engine does with a routed Unshelve: an operator-style unshelve.
                    unshelvedBy ??= $"Unshelve:{cmd.User}";
                    current = current with { Shelving = AlarmShelvingKind.Unshelved, UnshelveAtUtc = null };
                    rig.ProjectLater(native: false, current);
                    return;
                }
                if (cmd.Operation != "TimedShelve") return;
-                current = current with { Shelving = AlarmShelvingKind.Timed };
+                // The real engine's projection carries its expiry, so the re-projection keeps the SDK timer armed
                // for the real remaining time (without it, it would re-arm for MaxTimeShelved and never fire here).
                current = current with { Shelving = AlarmShelvingKind.Timed, UnshelveAtUtc = cmd.UnshelveAtUtc };
                rig.ProjectLater(native: false, current);
                var unshelveAt = cmd.UnshelveAtUtc!.Value;
                expiry = Task.Run(async () =>
                {
-                    var wait = unshelveAt - DateTime.UtcNow;
+                    var wait = unshelveAt + engineLag - DateTime.UtcNow;
                    if (wait > TimeSpan.Zero) await Task.Delay(wait);
-                    current = current with { Shelving = AlarmShelvingKind.Unshelved };
+                    if (current.Shelving != AlarmShelvingKind.Timed) return; // ApplyShelvingCheck: Timed only
                    unshelvedBy ??= "AutoUnshelve:system";
                    current = current with { Shelving = AlarmShelvingKind.Unshelved, UnshelveAtUtc = null };
                    rig.Project(native: false, current);
                });
            };
            // Count the SDK's own expiry-timer firings without changing what the production handler does.
            var condition = rig.Nm.TryGetAlarmCondition(ScriptedAlarmId)!;
            var sdkTimerFired = 0;
            var production = condition.OnTimedUnshelve!;
            condition.OnTimedUnshelve = (context, c) =>
            {
                Interlocked.Increment(ref sdkTimerFired);
                return production(context, c);
            };
            using var session = await rig.ConnectAsync(anonymous: false, ct);
            var events = new RecordedEvents();
            var subscription = await RecordedEvents.SubscribeAsync(session, events, ct);
@@ -315,6 +345,10 @@ public sealed class ConditionMethodEventTests
            rig.Project(false, current = current with { Message = "sentinel" });
            await RecordedEvents.WaitUntilAsync(() => events.Has(item, "sentinel"));
            Volatile.Read(ref sdkTimerFired).ShouldBeGreaterThanOrEqualTo(1, "the SDK's timed-unshelve timer fired");
            routed.ShouldNotContain("Unshelve", "the SDK timer must not route an operator-style Unshelve");
            unshelvedBy.ShouldBe("AutoUnshelve:system", "the engine's own expiry owns the unshelve");
            var shelving = events.ConditionEvents(item, "base").Select(e => e[RecordedEvents.ShelvingField].Value).ToList();
            shelving.ShouldBe(
                [ObjectIds.ShelvedStateMachineType_Unshelved, ObjectIds.ShelvedStateMachineType_TimedShelved,
@@ -268,19 +268,15 @@ public sealed class AlarmCommandRouterTests : IDisposable
        await host.DisposeAsync();
    }
-    /// <summary>OnTimedUnshelve fires with the SDK's system context (no session, no user identity) —
+    /// <summary>OnTimedUnshelve fires with the SDK's system context (no session, no user identity) when the
-    /// the real SDK path when a TimedShelve duration expires. The gate must NOT veto: the result must be
+    /// SDK's own TimedShelve timer expires. It must return Good and route NOTHING (review R1 M2, Gitea #597):
-    /// Good, the router must be invoked exactly once with Operation == "Unshelve", and no UnshelveAtUtc is
+    /// the scripted engine owns timed-shelve expiry (its shelving check records an "AutoUnshelve" by
-    /// carried.
+    /// "system"), and since #597 arms the SDK timer for the real remaining time, an "Unshelve" routed from
-    /// <para>
+    /// here pre-empted that check on every expiry, from both pair nodes, as an operator-style unshelve.
-    /// Regression (OpcUaServer-007): the routed command's <see cref="AlarmCommand.User"/> must be a
+    /// <para>Positive control: on the same condition and router, a client Unshelve (OnShelve with
-    /// NON-EMPTY system user (<c>"system"</c>). The engine's <c>Part9StateMachine.ApplyUnshelve</c> rejects
+    /// <c>shelving:false</c>) DOES route — so the empty capture is not a disconnected router.</para></summary>
    /// a null/whitespace user with <c>ArgumentException("User required.")</c>; that exception is swallowed
    /// downstream, so an empty user made the timed auto-unshelve SILENTLY no-op — a TimedShelve would never
    /// auto-expire. Asserting a non-empty user here pins the value the state machine requires.
    /// </para></summary>
    [Fact]
-    public async Task OnTimedUnshelve_with_system_context_returns_good_and_routes_unshelve()
+    public async Task OnTimedUnshelve_with_system_context_returns_good_and_routes_nothing()
    {
        var (host, server) = await BootAsync();
        var nm = server.NodeManager!;
@@ -299,14 +295,13 @@ public sealed class AlarmCommandRouterTests : IDisposable
        var result = condition.OnTimedUnshelve!(ctx, condition);
        result.ShouldBe(ServiceResult.Good);
-        captured.Count.ShouldBe(1);
+        captured.ShouldBeEmpty();
-        var cmd = captured[0];
+
-        cmd.AlarmId.ShouldBe("alm-tu");    // == ScriptedAlarmId / condition NodeId identifier
+        // Positive control: the router is live for a genuine operator Unshelve on the same condition.
-        cmd.Operation.ShouldBe("Unshelve");
+        var operatorCtx = SessionContext(server, "grace", OpcUaDataPlaneRoles.AlarmAck);
-        // System-initiated (no client principal), but the user MUST be non-empty so ApplyUnshelve accepts it.
+        condition.OnShelve!(operatorCtx, condition, shelving: false, oneShot: false, shelvingTime: 0.0)
-        cmd.User.ShouldNotBeNullOrWhiteSpace();
+            .ShouldBe(ServiceResult.Good);
-        cmd.User.ShouldBe("system");
+        captured.Select(c => c.Operation).ShouldBe(["Unshelve"]);
        cmd.UnshelveAtUtc.ShouldBeNull();
        await host.DisposeAsync();
    }
@@ -102,8 +102,14 @@ public sealed class DriverHostActorBootFromCacheTests : RuntimeActorTestBase
        // Barrier: diagnostics are served only after PreStart's boot work ran, so the absence below is
        // asserted after the boot path had its chance to Tell (positive control is the valid-bytes test).
-        AskDiagnostics(actor);
+        var snapshot = AskDiagnostics(actor);
        publishProbe.ExpectNoMsg(TimeSpan.FromMilliseconds(500));
        // Review R1 m4: nothing was applied, so the node is NOT running from cache and has NOT adopted the
        // cached revision (Stale with no configuration, retrying the ConfigDb) — the valid-bytes test above
        // is the control that both flip when the bytes are usable.
        snapshot.RunningFromCache.ShouldBeFalse();
        snapshot.CurrentRevision.ShouldBeNull();
    }
    [Fact]
@@ -111,6 +111,39 @@ public sealed class DriverHostActorFetchAndCacheBootstrapTests : RuntimeActorTes
        fetcher.Calls.ShouldBe(1);
    }
    /// <summary>
    ///     #601 / review R1 m4 — a cache row whose bytes are malformed applied nothing, so its revision must not
    ///     become current: the boot lands Steady-with-no-revision (as for an empty cache), and a re-dispatch of
    ///     the SAME revision really applies (fetches) instead of being acked by the "already at this rev"
    ///     short-circuit. <see cref="WithCachedArtifact_RestoresServedState_WithoutReadingCentralSql"/> is the
    ///     control: valid bytes do adopt the revision.
    /// </summary>
    [Fact]
    public void WithMalformedCachedBytes_TheRevisionIsNotAdopted_AndARedispatchOfItActuallyApplies()
    {
        var cached = new CachedDeploymentArtifact(
            DeploymentId.NewId().ToString(), CachedRev.Value,
            System.Text.Encoding.UTF8.GetBytes("{\"DriverInstances\": ["), DateTimeOffset.UtcNow.AddHours(-1));
        var fetcher = new RecordingFetcher(Artifact());
        var coordinator = CreateTestProbe();
        var actor = Sys.ActorOf(DriverHostActor.Props(
            new ThrowingDbFactory(), TestNode, coordinator.Ref,
            localRoles: new HashSet<string> { "driver" },
            deploymentArtifactCache: new StubArtifactCache(cached),
            fetchAndCacheMode: true,
            artifactFetcher: fetcher));
        var snapshot = AskDiagnostics(actor);
        snapshot.CurrentRevision.ShouldBeNull();
        snapshot.RunningFromCache.ShouldBeFalse();
        actor.Tell(new DispatchDeployment(DeploymentId.Parse(cached.DeploymentId), CachedRev, CorrelationId.NewId()));
        coordinator.ExpectMsg<ApplyAck>(TimeSpan.FromSeconds(5)).Outcome.ShouldBe(ApplyAckOutcome.Applied);
        fetcher.Calls.ShouldBe(1);
        AskDiagnostics(actor).CurrentRevision.ShouldBe(CachedRev);
    }
    private NodeDiagnosticsSnapshot AskDiagnostics(IActorRef actor)
    {
        var probe = CreateTestProbe();
@@ -0,0 +1,109 @@
 using Akka.Actor;
 using Shouldly;
 using Xunit;
 using ZB.MOM.WW.OtOpcUa.Commons.Messages.Redundancy;
 using ZB.MOM.WW.OtOpcUa.Commons.Types;
 using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
 using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
 using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
 namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
 /// <summary>
 ///     Review R1 m2 (Gitea #597) — the Primary data-plane verdict published to <see cref="IDriverPrimaryRoleSource"/>
 ///     (read by the native-ack gate and the Sparkplug primary-host claim) mirrors the host's own live gate from the
 ///     moment the host starts, not only from the first redundancy snapshot.
 /// </summary>
 /// <remarks>
 ///     <c>RedundancyRoleView</c> is seeded with the single-node verdict (allow). Before this fix nothing published to
 ///     it until a snapshot arrived, so on a pair both nodes accepted a native ack through the whole boot window while
 ///     <c>DriverHostActor</c> — which evaluates the driver member count live — denied the upstream route on the
 ///     non-Primary. Assertions are on what was PUBLISHED: the seed already reads <c>true</c>, so a current-value
 ///     assertion of the single-node case would pass with no publish at all.
 /// </remarks>
 public sealed class DriverHostActorPrimaryRoleSourceTests : RuntimeActorTestBase
 {
    private static readonly NodeId TestNode = NodeId.Parse("driver-primary-source-test");
    private static readonly TimeSpan Timeout = TimeSpan.FromSeconds(5);
    /// <summary>On a pair, an unknown role publishes DENY at start — before any snapshot. The later Primary snapshot
    /// is the positive control: the source is wired, and the deny was the verdict, not a missing publish.</summary>
    [Fact]
    public void A_paired_host_publishes_deny_before_the_first_snapshot()
    {
        var source = new RecordingPrimaryRoleSource();
        var host = SpawnHost(source, () => 2);
        AwaitAssert(() => source.Published.ShouldBe([false]), duration: Timeout);
        host.Tell(Snapshot(RedundancyRole.Primary));
        AwaitAssert(() => source.Published.ShouldBe([false, true]), duration: Timeout);
    }
    /// <summary>A single-driver node keeps the allow posture — it is published, not merely seeded.</summary>
    [Fact]
    public void A_single_node_host_publishes_allow_at_start()
    {
        var source = new RecordingPrimaryRoleSource();
        SpawnHost(source, () => 1);
        AwaitAssert(() => source.Published.ShouldBe([true]), duration: Timeout);
    }
    /// <summary>A driver peer coming Up while the role is still unknown flips the published verdict to deny, with no
    /// snapshot involved — the membership re-publish path.</summary>
    [Fact]
    public void A_peer_joining_before_any_snapshot_flips_the_published_verdict_to_deny()
    {
        var source = new RecordingPrimaryRoleSource();
        var members = 1;
        var host = SpawnHost(source, () => Volatile.Read(ref members));
        AwaitAssert(() => source.Published.ShouldBe([true]), duration: Timeout);
        Volatile.Write(ref members, 2);
        host.Tell(DriverHostActor.RepublishPrimaryVerdict.Instance);
        AwaitAssert(() => source.Published.ShouldBe([true, false]), duration: Timeout);
    }
    private IActorRef SpawnHost(IDriverPrimaryRoleSource source, Func<int> driverMemberCount) =>
        Sys.ActorOf(DriverHostActor.Props(
            NewInMemoryDbFactory(),
            TestNode,
            ackRouter: null,
            driverMemberCountProvider: driverMemberCount,
            primaryRoleSource: source));
    private static RedundancyStateChanged Snapshot(RedundancyRole role) =>
        new(
            [
                new NodeRedundancyState(TestNode, role,
                    IsClusterLeader: role == RedundancyRole.Primary,
                    IsDriverPrimary: role == RedundancyRole.Primary,
                    AsOfUtc: DateTime.UtcNow),
            ],
            CorrelationId.NewId());
    /// <summary>Remembers every published verdict, so tests assert on the actor's action, not on the seed.</summary>
    private sealed class RecordingPrimaryRoleSource : IDriverPrimaryRoleSource
    {
        private readonly List<bool> _published = [];
        private readonly Lock _gate = new();
        /// <summary>Every value published so far, in order.</summary>
        public IReadOnlyList<bool> Published
        {
            get { lock (_gate) return _published.ToArray(); }
        }
        public bool ShouldActAsPrimary
        {
            get { lock (_gate) return _published.Count == 0 || _published[^1]; }
        }
        public void PublishPrimaryRole(bool shouldActAsPrimary)
        {
            lock (_gate) _published.Add(shouldActAsPrimary);
        }
    }
 }
@@ -227,6 +227,77 @@ public sealed class DriverHostActorUnreadableArtifactTests : RuntimeActorTestBas
        replayed.ShouldContain("other-node-driver");
    }
    /// <summary>
    ///     #609 (review R1 M1) — the Resilience half of the eviction must SURVIVE the next resilience-publisher
    ///     tick. <c>DriverResilienceStatusPublisherService</c> re-emits <c>tracker.Snapshot()</c> into the hub
    ///     every interval, so evicting only the hub cache was undone within one tick: the host must also
    ///     release the instance's tracker state. The tick below is the publisher's own body
    ///     (<c>BuildMessages</c> → <c>hub.Emit</c>) minus the DPS publish, run against the REAL factory +
    ///     tracker the Host wires. The neighbour instance is the positive control: it is still emitted by
    ///     the same tick, so an absent drv-1 is not an empty tick.
    /// </summary>
    [Fact]
    public void Removing_a_driver_instance_releases_its_resilience_state_so_the_next_publisher_tick_omits_it()
    {
        var db = NewInMemoryDbFactory();
        var factory = new SubscribingDriverFactory("Modbus");
        var hub = new ZB.MOM.WW.OtOpcUa.Runtime.Telemetry.TelemetryLocalHub();
        var tracker = new ZB.MOM.WW.OtOpcUa.Core.Resilience.DriverResilienceStatusTracker();
        var invokerFactory = new ZB.MOM.WW.OtOpcUa.Core.Resilience.DriverCapabilityInvokerFactory(
            new ZB.MOM.WW.OtOpcUa.Core.Resilience.DriverResiliencePipelineBuilder(statusTracker: tracker),
            tracker,
            _ => ZB.MOM.WW.OtOpcUa.Core.Abstractions.DriverTier.A);
        var coordinator = CreateTestProbe();
        var actor = Sys.ActorOf(DriverHostActor.Props(
            db, TestNode, coordinator.Ref,
            driverFactory: factory,
            localRoles: new HashSet<string> { "driver" },
            invokerFactory: invokerFactory,
            telemetryHub: hub));
        actor.Tell(new DispatchDeployment(SeedV3Deployment(db, RevA), RevA, CorrelationId.NewId()));
        coordinator.ExpectMsg<ApplyAck>(Timeout).Outcome.ShouldBe(ApplyAckOutcome.Applied);
        // Resilience counters for the running driver and for an instance this host does not own.
        tracker.RecordBreakerOpen("drv-1", "dev1", DateTime.UtcNow);
        tracker.RecordFailure("neighbour-drv", "dev9", DateTime.UtcNow);
        PublisherTick(tracker, hub);
        ReplayedResilienceIds(hub).ShouldContain("drv-1"); // control: drv-1 is primed before the removal
        actor.Tell(new DispatchDeployment(SeedDriverlessDeployment(db, RevB), RevB, CorrelationId.NewId()));
        coordinator.ExpectMsg<ApplyAck>(Timeout).Outcome.ShouldBe(ApplyAckOutcome.Applied);
        AwaitAssert(
            () => AskDiagnostics(actor).Drivers.Select(d => d.Name).ShouldNotContain("drv-1"),
            duration: Timeout);
        // The publisher's next tick, after the removal.
        PublisherTick(tracker, hub);
        var replayed = ReplayedResilienceIds(hub);
        replayed.ShouldNotContain("drv-1");
        replayed.ShouldContain("neighbour-drv");
        static void PublisherTick(
            ZB.MOM.WW.OtOpcUa.Core.Resilience.DriverResilienceStatusTracker t,
            ZB.MOM.WW.OtOpcUa.Runtime.Telemetry.TelemetryLocalHub h)
        {
            var now = DateTime.UtcNow;
            foreach (var e in t.Snapshot())
                h.Emit(new ZB.MOM.WW.OtOpcUa.Runtime.Telemetry.TelemetryItem.Resilience(new(
                    e.DriverInstanceId, e.HostName, e.Snapshot.IsBreakerOpen, e.Snapshot.ConsecutiveFailures,
                    e.Snapshot.CurrentInFlight, e.Snapshot.LastBreakerOpenUtc, e.Snapshot.LastSampledUtc, now)));
        }
        static List<string> ReplayedResilienceIds(ZB.MOM.WW.OtOpcUa.Runtime.Telemetry.TelemetryLocalHub h)
        {
            using var sub = h.Subscribe(64);
            var ids = new List<string>();
            while (sub.Reader.TryRead(out var i))
                if (i is ZB.MOM.WW.OtOpcUa.Runtime.Telemetry.TelemetryItem.Resilience r) ids.Add(r.E.DriverInstanceId);
            return ids;
        }
    }
    /// <summary>Spawns the host with the subscribing factory, dispatches <paramref name="deploymentId"/> and
    /// waits for the Applied ACK so the child + the initial SubscribeBulk pass are in place.</summary>
    private (IActorRef Actor, Akka.TestKit.TestProbe Coordinator) SpawnHostAndApply(
@@ -109,7 +109,8 @@ public sealed class AlarmHistorianRegistrationTests
    /// <remarks>
    ///     <para>
    ///         Standing down is only safe because a peer holds the same rows and will send them
-    ///         instead. Without <c>LocalDb:Replication:PeerAddress</c> there is no such peer: the rows
+    ///         instead. Without an authenticated sync path (<c>ApiKey</c> plus a dial or a listener —
    ///         <c>LocalDbReplicationPolicy.SharesRowsWithPeer</c>) there is no such peer: the rows
    ///         are here and nowhere else, so any deferral means nobody ever delivers them.
    ///     </para>
    ///     <para>
@@ -124,14 +125,18 @@ public sealed class AlarmHistorianRegistrationTests
    // Neither key: an unpaired node. Its rows exist nowhere else.
    [InlineData(null, null, null, true)]
    // The dialing half of a pair — gated, because its partner holds the same rows.
-    [InlineData("http://peer:9001", null, null, false)]
+    [InlineData("http://peer:9001", null, "k", false)]
    // The LISTENING half. Same shared queue, but it never dials, so a PeerAddress-only test would
-    // wrongly class it as unpaired and leave it permanently ungated. It is identified by the shared
+    // wrongly class it as unpaired and leave it permanently ungated. Key + listen port = a peer.
    // ApiKey (the same predicate LocalDbSetup registers tables on), not by its listen port.
    [InlineData(null, "9001", "k", false)]
    // A listen port with NO key: the fail-closed passive endpoint rejects every inbound stream, so the
    // queue is shared with nobody and the node must keep draining (#603: this used to be gated).
    [InlineData(null, "9001", null, true)]
    // A dial with NO key: the peer's fail-closed endpoint rejects it — shares nothing, keeps draining.
    [InlineData("http://peer:9001", null, null, true)]
    // A key with NEITHER path (review R1 m3: e.g. a fleet-wide secret on an unpaired node) — no listener,
    // no dial, so no peer holds these rows. Registered for capture, but must keep draining.
    [InlineData(null, null, "k", true)]
    public async Task Only_an_unreplicated_node_ignores_the_role_view(
        string? peerAddress, string? syncListenPort, string? apiKey, bool expectDrain)
    {
@@ -4,29 +4,64 @@ using Xunit;
 namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Historian;
-/// <summary>#603: the single "is this node's LocalDb replicated" predicate shared by table registration and the alarm drain gate.</summary>
+/// <summary>
 ///     #603 + review R1 m3: the two "is this node's LocalDb replicated" predicates — permissive for table
 ///     registration (<see cref="LocalDbReplicationPolicy.CapturesChanges"/>), strict for the alarm drain gate
 ///     (<see cref="LocalDbReplicationPolicy.SharesRowsWithPeer"/>).
 /// </summary>
 public sealed class LocalDbReplicationPolicyTests
 {
    private const string Peer = "LocalDb:Replication:PeerAddress";
    private const string Key = "LocalDb:Replication:ApiKey";
    private const string Port = "LocalDb:SyncListenPort";
    private static IConfiguration Config(params (string Key, string? Value)[] pairs) =>
        new ConfigurationBuilder()
            .AddInMemoryCollection(pairs.Select(p => new KeyValuePair<string, string?>(p.Key, p.Value)))
            .Build();
-    /// <summary>PeerAddress OR ApiKey counts; the listen port alone and blank values do not.</summary>
+    /// <summary>Registration: PeerAddress OR ApiKey counts; the listen port alone and blank values do not.</summary>
    [Theory]
-    [InlineData("LocalDb:Replication:PeerAddress", "http://peer:9001", true)]
+    [InlineData(Peer, "http://peer:9001", true)]
-    [InlineData("LocalDb:Replication:ApiKey", "k", true)]
+    [InlineData(Key, "k", true)]
-    [InlineData("LocalDb:SyncListenPort", "9001", false)]
+    [InlineData(Port, "9001", false)]
-    [InlineData("LocalDb:Replication:ApiKey", "  ", false)]
+    [InlineData(Key, "  ", false)]
-    public void IsConfigured_follows_the_PeerAddress_or_ApiKey_rule(string key, string value, bool expected)
+    public void CapturesChanges_follows_the_PeerAddress_or_ApiKey_rule(string key, string value, bool expected)
    {
-        LocalDbReplicationPolicy.IsConfigured(Config((key, value))).ShouldBe(expected);
+        LocalDbReplicationPolicy.CapturesChanges(Config((key, value))).ShouldBe(expected);
    }
-    /// <summary>No replication keys at all is an unpaired node.</summary>
+    /// <summary>Drain gate: ApiKey AND (PeerAddress OR SyncListenPort &gt; 0).</summary>
-    [Fact]
+    [Theory]
-    public void IsConfigured_is_false_for_an_empty_configuration()
+    [InlineData("http://peer:9001", "k", null, true)]   // dialling half of a pair
    [InlineData(null, "k", "9001", true)]               // listening half of a pair
    [InlineData("http://peer:9001", "k", "9001", true)] // both
    [InlineData(null, "k", null, false)]                // key, no path: shares nothing (review R1 m3)
    [InlineData(null, "k", "0", false)]                 // port 0 = no listener
    [InlineData(null, "k", "nope", false)]              // unparseable port = no listener
    [InlineData("http://peer:9001", null, "9001", false)] // paths, no key: both auth directions fail closed
    [InlineData(null, "  ", "9001", false)]             // blank key
    public void SharesRowsWithPeer_needs_a_key_and_a_sync_path(string? peer, string? key, string? port, bool expected)
    {
-        LocalDbReplicationPolicy.IsConfigured(Config()).ShouldBeFalse();
+        LocalDbReplicationPolicy.SharesRowsWithPeer(Config((Peer, peer), (Key, key), (Port, port))).ShouldBe(expected);
    }
    /// <summary>Every node the gate treats as sharing rows also captures — never a gated node with no capture.</summary>
    [Theory]
    [InlineData("http://peer:9001", "k", null)]
    [InlineData(null, "k", "9001")]
    public void A_node_sharing_rows_always_captures(string? peer, string? key, string? port)
    {
        var config = Config((Peer, peer), (Key, key), (Port, port));
        LocalDbReplicationPolicy.SharesRowsWithPeer(config).ShouldBeTrue();
        LocalDbReplicationPolicy.CapturesChanges(config).ShouldBeTrue();
    }
    /// <summary>No replication keys at all is an unpaired node under both predicates.</summary>
    [Fact]
    public void Both_predicates_are_false_for_an_empty_configuration()
    {
        LocalDbReplicationPolicy.CapturesChanges(Config()).ShouldBeFalse();
        LocalDbReplicationPolicy.SharesRowsWithPeer(Config()).ShouldBeFalse();
    }
 }