Phase 6.1 Stream A.3 — wrap all 4 HistoryRead dispatch paths through CapabilityInvoker

Per Stream A.3 coverage goal, every IHistoryProvider method on the server
dispatch surface routes through the invoker with DriverCapability.HistoryRead:
- HistoryReadRaw  (line 487)
- HistoryReadProcessed  (line 551)
- HistoryReadAtTime  (line 608)
- HistoryReadEvents  (line 665)

Each gets timeout + per-(driver, host) circuit breaker + the default Tier
retry policy (Tier A default: 2 retries at 30s timeout). Inner driver
GetAwaiter().GetResult() pattern preserved because the OPC UA stack's
HistoryRead hook is sync-returning-void — see CustomNodeManager2.

With Read, Write, and HistoryRead wrapped, Stream A's invoker-coverage
compliance check passes for the dispatch surfaces that live in
DriverNodeManager. Subscribe / AlarmSubscribe / AlarmAcknowledge sit behind
push-based subscription plumbing (driver → OPC UA event layer) rather than
server-pull dispatch, so they're wrapped in the driver-to-server glue rather
than in DriverNodeManager — deferred to the follow-up PR that wires the
remaining capability surfaces per the final Roslyn-analyzer-enforced coverage
map.

Full solution dotnet test: 948 passing. Pre-existing Client.CLI Subscribe
flake unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/v2/implementation/phase-6-1-resilience-and-observability.md

@@ -0,0 +1,149 @@
+# Phase 6.1 — Resilience & Observability Runtime
+
+> **Status**: DRAFT — implementation plan for a cross-cutting phase that was never formalised. The v2 `plan.md` specifies Polly, Tier A/B/C protections, structured logging, and local-cache fallback by decision; none are wired end-to-end.
+>
+> **Branch**: `v2/phase-6-1-resilience-observability`
+> **Estimated duration**: 3 weeks
+> **Predecessor**: Phase 5 (drivers) — partial; S7 + OPC UA Client shipped, AB/TwinCAT/FOCAS paused
+> **Successor**: Phase 6.2 (Authorization runtime)
+
+## Phase Objective
+
+Land the cross-cutting runtime protections + operability features that `plan.md` + `driver-stability.md` specify by decision but that no driver-phase actually wires. End-state: every driver goes through the same Polly resilience layer, health endpoints render the live driver fleet, structured logs carry per-request correlation IDs, and the config substrate survives a central DB outage via a LiteDB local cache.
+
+Closes these gaps flagged in the 2026-04-19 audit:
+
+1. Polly v8 resilience pipelines wired to every `IDriver` capability (no-op per-driver today; Galaxy has a hand-rolled `CircuitBreaker` only).
+2. Tier A/B/C enforcement at runtime — `driver-stability.md` §2–4 and decisions #63–73 define memory watchdog, bounded queues, scheduled recycle, wedge detection; `MemoryWatchdog` exists only inside `Driver.Galaxy.Host`.
+3. Health endpoints (`/healthz`, `/readyz`) on `OtOpcUa.Server`.
+4. Structured Serilog with per-request correlation IDs (driver instance, OPC UA session, IPC call).
+5. LiteDB local cache + Polly retry + fallback on central-DB outage (decision #36).
+
+## Scope — What Changes
+
+| Concern | Change |
+|---------|--------|
+| `Core` → new `Core.Resilience` sub-namespace | Shared Polly pipeline builder (`DriverResiliencePipelines`). **Pipeline key = `(DriverInstanceId, HostName)`** so one dead PLC behind a multi-device driver doesn't open the breaker for healthy siblings (decision #35 per-device isolation). **Per-capability policy** — Read / HistoryRead / Discover / Probe / Alarm get retries; **Write does NOT** unless `[WriteIdempotent]` on the tag definition (decisions #44-45). |
+| Every capability-interface consumer in the server | Wrap `IReadable.ReadAsync`, `IWritable.WriteAsync`, `ITagDiscovery.DiscoverAsync`, `ISubscribable.SubscribeAsync/UnsubscribeAsync`, `IHostConnectivityProbe` probe loop, `IAlarmSource.SubscribeAlarmsAsync/AcknowledgeAsync`, `IHistoryProvider.ReadRawAsync/ReadProcessedAsync/ReadAtTimeAsync/ReadEventsAsync`. Composition: timeout → (retry when capability supports) → circuit breaker → bulkhead. |
+| `Core.Abstractions` → new `WriteIdempotentAttribute` | Marker on `ModbusTagDefinition` / `S7TagDefinition` / `OpcUaClientDriver` tag rows; opts that tag into auto-retry on Write. Absence = no retry, per spec. |
+| `Core` → new `Core.Stability` sub-namespace — **split** | Two separate subsystems: (a) **`MemoryTracking`** runs all tiers; captures baseline (median of first 5 min `GetMemoryFootprint` samples) + applies the hybrid rule `soft = max(multiplier × baseline, baseline + floor)`; soft breach logs + surfaces to Admin; never kills. (b) **`MemoryRecycle`** (Tier C only — requires out-of-process topology) handles hard-breach recycle via the Proxy-side supervisor. Tier A/B overrun escalates to Tier C promotion ticket, not auto-kill. |
+| `ScheduledRecycleScheduler` | Tier C only per decisions #73-74. Weekly/time-of-day recycle via Proxy supervisor. Tier A/B opt-in recycle lands in a future phase together with a Tier-C-escalation workflow. |
+| `WedgeDetector` | **Demand-aware**: flips a driver to Faulted only when `(hasPendingWork AND noProgressIn > threshold)`. `hasPendingWork` derives from non-zero Polly bulkhead depth OR ≥1 active MonitoredItem OR ≥1 queued historian read. Idle + subscription-only drivers stay Healthy. |
+| `DriverTypeRegistry` | Each driver type registers its `DriverTier` {A, B, C}. Tier C drivers must advertise their out-of-process topology; the registry enforces invariants (Tier C has a `Proxy` + `Host` pair). |
+| `Driver.Galaxy.Proxy/Supervisor/` | **Retains** existing `CircuitBreaker` + `Backoff` — they guard IPC respawn (decision #68), different concern from the per-call Polly layer. Only `HeartbeatMonitor` is referenced downstream (IPC liveness). |
+| `OtOpcUa.Server` → Minimal API endpoints on `http://+:4841` | `/healthz` = process alive + (config DB reachable OR `UsingStaleConfig=true`). `/readyz` = ANDed driver health; state-machine per `DriverState`: `Unknown`/`Initializing` → 503, `Healthy` → 200, `Degraded` → 200 + `{degradedDrivers: [...]}` in body, `Faulted` → 503. JSON body always reports per-instance detail. |
+| Serilog configuration | Centralize enrichers in `OtOpcUa.Server/Observability/LogContextEnricher.cs`. Every capability call runs inside a `LogContext.PushProperty` scope with {DriverInstanceId, DriverType, CapabilityName, CorrelationId (UA RequestHandle or internal GUID)}. Sink config stays rolling-file per CLAUDE.md; JSON sink added alongside plain-text (switchable via `Serilog:WriteJson` appsetting). |
+| `Configuration` project | Add `LiteDbConfigCache` adapter. **Generation-sealed snapshots**: `sp_PublishGeneration` writes `<cache-root>/<cluster>/<generationId>.db` as a read-only sealed file. Reads serve the last-known-sealed generation; mixed-generation reads are impossible. Write path bypasses cache + fails hard on DB outage. Pipeline: timeout (2 s) → retry (3×, jittered) → fallback-to-sealed-snapshot. |
+| `DriverHostStatus` vs. `DriverInstanceResilienceStatus` | New separate entity `DriverInstanceResilienceStatus { DriverInstanceId, HostName, LastCircuitBreakerOpenUtc, ConsecutiveFailures, CurrentBulkheadDepth, LastRecycleUtc, BaselineFootprintBytes }`. `DriverHostStatus` keeps per-host connectivity only; Admin `/hosts` joins both for display. |
+
+## Scope — What Does NOT Change
+
+| Item | Reason |
+|------|--------|
+| Driver wire protocols | Resilience is a server-side wrapper; individual drivers don't see Polly. Their existing retry logic (ModbusTcpTransport reconnect, SessionReconnectHandler) stays in place as inner layers. |
+| Config DB schema | LiteDB cache is a read-only mirror; no new central tables except `DriverHostStatus` column additions. |
+| OPC UA wire behavior visible to clients | Health endpoints live on a separate HTTP port (4841 by convention); the OPC UA server on 4840 is unaffected. |
+| The four 2026-04-13 Galaxy stability findings | Already closed in Phase 2. Phase 6.1 *generalises* the pattern, doesn't re-fix Galaxy. |
+| Driver-layer SafeHandle usage | Existing Galaxy `SafeMxAccessHandle` + Modbus `TcpClient` disposal stay — they're driver-internal, not part of the cross-cutting layer. |
+
+## Entry Gate Checklist
+
+- [ ] Phases 0–5 exit gates cleared (or explicitly deferred with task reference)
+- [ ] `driver-stability.md` §2–4 re-read; decisions #63–73 + #34–36 re-skimmed
+- [ ] Polly v8 NuGet available (`Microsoft.Extensions.Resilience` + `Polly.Core`) — verify package restore before task breakdown
+- [ ] LiteDB 5.x NuGet confirmed MIT + actively maintained
+- [ ] Existing drivers catalogued: Galaxy.Proxy, Modbus, S7, OpcUaClient — confirm test counts baseline so the resilience layer doesn't regress any
+- [ ] Serilog configuration inventory: locate every `Log.ForContext` call site that will need `LogContext` rewrap
+- [ ] Admin `/hosts` page's current `DriverHostStatus` consumption reviewed so the schema extensions don't break it
+
+## Task Breakdown
+
+### Stream A — Resilience layer (1 week)
+
+1. **A.1** Add `Polly.Core` + `Microsoft.Extensions.Resilience` to `Core`. Build `DriverResiliencePipelineBuilder` — key on `(DriverInstanceId, HostName)`; composes Timeout → (Retry when the capability allows it; skipped for Write unless `[WriteIdempotent]`) → CircuitBreaker → Bulkhead. Per-capability policy map documented in `DriverResilienceOptions.CapabilityPolicies`.
+2. **A.2** `DriverResilienceOptions` record bound from `DriverInstance.ResilienceConfig` JSON column (new nullable). **Per-tier × per-capability** defaults: Tier A (OpcUaClient, S7) Read 3 retries/2 s/5-failure-breaker, Write 0 retries/2 s/5-failure-breaker; Tier B (Modbus) Read 3/4 s/5, Write 0/4 s/5; Tier C (Galaxy) Read 1 retry/10 s/no-kill, Write 0/10 s/no-kill. Idempotent writes can opt into Read-shaped retry via the attribute.
+3. **A.3** `CapabilityInvoker<TCapability, TResult>` wraps every method on the capability interfaces (`IReadable.ReadAsync`, `IWritable.WriteAsync`, `ITagDiscovery.DiscoverAsync`, `ISubscribable.SubscribeAsync/UnsubscribeAsync`, `IHostConnectivityProbe` probe loop, `IAlarmSource.SubscribeAlarmsAsync/AcknowledgeAsync`, `IHistoryProvider.ReadRawAsync/ReadProcessedAsync/ReadAtTimeAsync/ReadEventsAsync`). Existing server-side dispatch routes through it.
+4. **A.4** **Retain** `Driver.Galaxy.Proxy/Supervisor/CircuitBreaker.cs` + `Backoff.cs` — they guard IPC process respawn (decision #68), orthogonal to the per-call Polly layer. Only `HeartbeatMonitor` is consumed outside the supervisor.
+5. **A.5** Unit tests: per-policy, per-composition. Negative integration tests: (a) Modbus FlakeyTransport fails 5× on Read, succeeds 6th — invoker surfaces success; (b) Modbus FlakeyTransport fails 1× on Write with `[WriteIdempotent]=false` — invoker surfaces failure without retry (no duplicate pulse); (c) Modbus FlakeyTransport fails 1× on Write with `[WriteIdempotent]=true` — invoker retries. Bench: no-op overhead < 1%.
+6. **A.6** `WriteIdempotentAttribute` in `Core.Abstractions`. Modbus/S7/OpcUaClient tag-definition records pick it up; invoker reads via reflection once at driver init.
+
+### Stream B — Tier A/B/C stability runtime — split into MemoryTracking + MemoryRecycle (1 week)
+
+1. **B.1** `Core.Abstractions` → `DriverTier` enum {A, B, C}. Extend `DriverTypeRegistry` to require `DriverTier` at registration. Existing driver types stamped (Galaxy = C, Modbus = B, S7 = B, OpcUaClient = A).
+2. **B.2** **`MemoryTracking`** (all tiers) lifted from `Driver.Galaxy.Host/MemoryWatchdog.cs`. Captures `BaselineFootprintBytes` as the median of first 5 min of `IDriver.GetMemoryFootprint()` samples post-`InitializeAsync`. Applies **decision #70 hybrid formula**: `soft = max(multiplier × baseline, baseline + floor)`; Tier A multiplier=3, floor=50 MB; Tier B multiplier=3, floor=100 MB; Tier C multiplier=2, floor=500 MB. Soft breach → log + `DriverInstanceResilienceStatus.CurrentFootprint` tick; never kills. Hard = 2 × soft.
+3. **B.3** **`MemoryRecycle`** (Tier C only per decisions #73-74). Hard-breach on a Tier C driver triggers `ScheduledRecycleScheduler.RequestRecycleNow(driverInstanceId)`; scheduler proxies to `Driver.Galaxy.Proxy/Supervisor/` which restarts the Host process. Tier A/B hard-breach logs a promotion-to-Tier-C recommendation; **never auto-kills** the in-process driver.
+4. **B.4** **`ScheduledRecycleScheduler`** per decision #67: Tier C driver instances opt-in to a weekly recycle at a configured cron. Tier A/B scheduled recycle deferred to a later phase paired with Tier-C escalation.
+5. **B.5** **`WedgeDetector`** demand-aware: `if (state==Healthy && hasPendingWork && noProgressIn > WedgeThreshold) → force ReinitializeAsync`. `hasPendingWork` = (bulkhead depth > 0) OR (active monitored items > 0) OR (queued historian-read count > 0). `WedgeThreshold` default 5 × PublishingInterval, min 60 s. Idle driver stays Healthy.
+6. **B.6** Tests: tracking unit tests drive synthetic allocation against a fake `GetMemoryFootprint`; recycle tests use a mock supervisor; wedge tests include the false-fault cases — idle subscriber, slow historian backfill, write-only burst.
+
+### Stream C — Health endpoints + structured logging (4 days)
+
+1. **C.1** `OtOpcUa.Server/Observability/HealthEndpoints.cs` — Minimal API on a second Kestrel binding (default `http://+:4841`). `/healthz` reports process uptime + config-DB reachability (or cache-warm). `/readyz` enumerates `DriverInstance` rows + reports each driver's `DriverHealth.State`; returns 503 if ANY driver is Faulted. JSON body per `docs/v2/acl-design.md` §"Operator Dashboards" shape.
+2. **C.2** `LogContextEnricher` installed at Serilog config time. Every driver-capability call site wraps its body in `using (LogContext.PushProperty("DriverInstanceId", id)) using (LogContext.PushProperty("CorrelationId", correlationId))`. Correlation IDs: reuse OPC UA `RequestHeader.RequestHandle` when in-flight; otherwise generate `Guid.NewGuid().ToString("N")[..12]`.
+3. **C.3** Add JSON-formatted Serilog sink alongside the existing rolling-file plain-text sink so SIEMs (Splunk, Datadog) can ingest without a regex parser. Sink switchable via `Serilog:WriteJson` appsetting.
+4. **C.4** Integration test: boot server, issue Modbus read, assert log line contains `DriverInstanceId` + `CorrelationId` structured fields.
+
+### Stream D — Config DB LiteDB fallback — generation-sealed snapshots (1 week)
+
+1. **D.1** `LiteDbConfigCache` adapter backed by **sealed generation snapshots**: each successful `sp_PublishGeneration` writes `<cache-root>/<clusterId>/<generationId>.db` as read-only after commit. The adapter maintains a `CurrentSealedGenerationId` pointer updated atomically on successful publish. Mixed-generation reads are **impossible** — every read served from the cache serves one coherent sealed generation.
+2. **D.2** Write-path queries (draft save, publish) bypass the cache entirely and fail hard on DB outage. Read-path queries (DriverInstance enumeration, LdapGroupRoleMapping, cluster + namespace metadata) go through the pipeline: timeout 2 s → retry 3× jittered → fallback to the current sealed snapshot.
+3. **D.3** `UsingStaleConfig` flag flips true when a read fell back to the sealed snapshot; cleared on the next successful DB round-trip. Surfaced on `/healthz` body and Admin `/hosts`.
+4. **D.4** Tests: (a) SQL-container kill mid-operation — read returns sealed snapshot, `UsingStaleConfig=true`, driver stays Healthy; (b) mixed-generation guard — attempt to serve partial generation by corrupting a snapshot file mid-read → adapter fails closed rather than serving mixed data; (c) first-boot-no-snapshot case — adapter refuses to start, driver fails `InitializeAsync` with a clear config-DB-required error.
+
+### Stream E — Admin `/hosts` page refresh (3 days)
+
+1. **E.1** Extend `DriverHostStatus` schema with Stream A resilience columns. Generate EF migration.
+2. **E.2** `Admin/FleetStatusHub` SignalR hub pushes `LastCircuitBreakerOpenUtc` + `CurrentBulkheadDepth` + `LastRecycleUtc` on change.
+3. **E.3** `/hosts` Blazor page renders new columns; red badge if `ConsecutiveFailures > breakerThreshold / 2`.
+
+## Compliance Checks (run at exit gate)
+
+- [ ] **Invoker coverage**: every method on `IReadable` / `IWritable` / `ITagDiscovery` / `ISubscribable` / `IHostConnectivityProbe` / `IAlarmSource` / `IHistoryProvider` in the server dispatch layer routes through `CapabilityInvoker`. Enforce via a Roslyn analyzer (error-level; warning-first is rejected — the compliance check is the gate).
+- [ ] **Write-retry guard**: writes without `[WriteIdempotent]` never get retried. Unit-test the invoker path asserts zero retry attempts.
+- [ ] **Pipeline isolation**: pipeline key is `(DriverInstanceId, HostName)`. Integration test with two Modbus hosts under one instance — failing host A does not open the breaker for host B.
+- [ ] **Tier registry**: every driver type registered in `DriverTypeRegistry` has a non-null `Tier`. Unit test walks the registry + asserts no gaps. Tier C registrations must declare their out-of-process topology.
+- [ ] **MemoryTracking never kills**: soft/hard breach tests on a Tier A/B driver log + surface without terminating the process.
+- [ ] **MemoryRecycle Tier C only**: hard breach on a Tier A driver never invokes the supervisor; on Tier C it does.
+- [ ] **Wedge demand-aware**: test suite includes idle-subscription-only, slow-historian-backfill, and write-only-burst cases — driver stays Healthy.
+- [ ] **Galaxy supervisor preserved**: `Driver.Galaxy.Proxy/Supervisor/CircuitBreaker.cs` + `Backoff.cs` still present + still invoked on Host crash.
+- [ ] **Health state machine**: `/healthz` + `/readyz` respond within 500 ms for every `DriverState`; state-machine table in this doc drives the test matrix.
+- [ ] **Structured log**: CI grep asserts at least one log line per capability call has `"DriverInstanceId"` + `"CorrelationId"` JSON fields.
+- [ ] **Generation-sealed cache**: integration tests cover (a) SQL-kill mid-operation serves last-sealed snapshot; (b) mixed-generation corruption fails closed; (c) first-boot no-snapshot + DB-down → `InitializeAsync` fails with clear error.
+- [ ] No regression in existing test suites — `dotnet test ZB.MOM.WW.OtOpcUa.slnx` count equal-or-greater than pre-Phase-6.1 baseline.
+
+## Risks and Mitigations
+
+| Risk | Likelihood | Impact | Mitigation |
+|------|:----------:|:------:|------------|
+| Polly pipeline adds per-request latency on hot path | Medium | Medium | Benchmark Stream A.5 before merging; 1 % overhead budget; inline hot path short-circuits when retry count = 0 |
+| LiteDB cache diverges from central DB | Medium | High | Stale-data banner in Admin UI; `UsingStaleConfig` flag surfaced on `/readyz`; cache refresh on every successful DB round-trip; 24-hour synthetic warning |
+| Tier watchdog false-positive-kills a legitimate batch load | Low | High | Soft/hard threshold split; soft only logs; hard triggers recycle; thresholds configurable per-instance |
+| Wedge detector races with slow-but-healthy drivers | Medium | High | Minimum 60 s threshold; detector only activates if driver claims `Healthy`; add circuit-breaker feedback so rapid oscillation trips instead of thrashing |
+| Roslyn analyzer breaks external driver authors | Low | Medium | Release analyzer as warning-level initially; upgrade to error in Phase 6.1+1 after one release cycle |
+
+## Completion Checklist
+
+- [ ] Stream A: Polly shared pipeline + per-tier defaults + driver-capability invoker + tests
+- [ ] Stream B: Tier registry + generalised watchdog + scheduled recycle + wedge detector
+- [ ] Stream C: `/healthz` + `/readyz` + structured logging + JSON Serilog sink
+- [ ] Stream D: LiteDB cache + Polly fallback in Configuration
+- [ ] Stream E: Admin `/hosts` page refresh
+- [ ] Cross-cutting: `phase-6-1-compliance.ps1` exits 0; full solution `dotnet test` passes; exit-gate doc recorded
+
+## Adversarial Review — 2026-04-19 (Codex, thread `019da489-e317-7aa1-ab1f-6335e0be2447`)
+
+Plan substantially rewritten before implementation to address these findings. Each entry: severity · verdict · adjustment.
+
+1. **Crit · ACCEPT** — Auto-retry collides with decisions #44/#45 (no auto-write-retry; opt-in via `WriteIdempotent` + CAS). Pipeline now **capability-specific**: Read/HistoryRead/Discover/Probe/Alarm-subscribe all get retries; **Write does not** unless the tag metadata carries `WriteIdempotent=true`. New `WriteIdempotentAttribute` surfaces on `ModbusTagDefinition` / `S7TagDefinition` / etc.
+2. **Crit · ACCEPT** — "One pipeline per driver instance" breaks decision #35's per-device isolation. **Change**: pipeline key is `(DriverInstanceId, HostName)` not just `DriverInstanceId`. One dead PLC behind a multi-device Modbus driver no longer opens the breaker for healthy siblings.
+3. **Crit · ACCEPT** — Memory watchdog + scheduled recycle at Tier A/B breaches decisions #73/#74 (process-kill protections are Tier-C-only). **Change**: Stream B splits into two — `MemoryTracking` (all tiers, soft/hard thresholds log + surface to Admin `/hosts`; never kills) and `MemoryRecycle` (Tier C only, requires out-of-process topology). Tier A/B overrun paths escalate to Tier C via a future PR, not auto-kill.
+4. **High · ACCEPT** — Removing Galaxy's hand-rolled `CircuitBreaker` drops decision #68 host-supervision crash-loop protection. **Change**: keep `Driver.Galaxy.Proxy/Supervisor/CircuitBreaker.cs` + `Backoff.cs` — they guard the IPC *process* re-spawn, not the per-call data path. Data-path Polly is an orthogonal layer.
+5. **High · ACCEPT** — Roslyn analyzer targeting `IDriver` misses the hot paths (`IReadable.ReadAsync`, `IWritable.WriteAsync`, `ISubscribable.SubscribeAsync` etc.). **Change**: analyzer rule now matches every method on the capability interfaces; compliance doc enumerates the full call-site list.
+6. **High · ACCEPT** — `/healthz` + `/readyz` under-specified for degraded-running. **Change**: add a state-matrix sub-section explicitly covering `Unknown` (pre-init: `/readyz` 503), `Initializing` (503), `Healthy` (200), `Degraded` (200 with JSON body flagging the degraded driver; `/readyz` is OR across drivers), `Faulted` (503), plus cached-config-serving (`/healthz` returns 200 + `UsingStaleConfig: true` in JSON body).
+7. **High · ACCEPT** — `WedgeDetector` based on "no successful Read" false-fires on write-only subscriptions + idle systems. **Change**: wedge criteria now `(hasPendingWork AND noProgressIn > threshold)` where `hasPendingWork` comes from the Polly bulkhead depth + active MonitoredItem count. Idle driver stays Healthy.
+8. **High · ACCEPT** — LiteDB cache serving mixed-generation reads breaks publish atomicity. **Change**: cache is snapshot-per-generation. Each published generation writes a sealed snapshot into `<cache-root>/<cluster>/<generationId>.db`; reads serve the last-known-sealed generation and never mix. Central DB outage during a *publish* means that publish fails (write path doesn't use cache); reads continue from the prior sealed snapshot.
+9. **Med · ACCEPT** — `DriverHostStatus` schema conflates per-host connectivity with per-driver-instance resilience counters. **Change**: new `DriverInstanceResilienceStatus` table separate from `DriverHostStatus`. Admin `/hosts` joins both for display.
+10. **Med · ACCEPT** — Compliance says analyzer-error; risks say analyzer-warning. **Change**: phase 6.1 ships at **error** level (this phase is the gate); warning-mode option removed.
+11. **Med · ACCEPT** — Hardcoded per-tier MB bands ignore decision #70's `max(multiplier × baseline, baseline + floor)` formula with observed-baseline capture. **Change**: watchdog captures baseline at post-init plateau (median of first 5 min GetMemoryFootprint samples) + applies the hybrid formula. Tier constants now encode the multiplier + floor, not raw MB.
+12. **Med · ACCEPT** — Tests mostly cover happy path. **Change**: Stream A.5 adds negative tests for duplicate-write-replay-under-timeout; Stream B.5 adds false-wedge-on-idle-subscription + false-wedge-on-slow-historic-backfill; Stream D.4 adds mixed-generation cache test + corrupt-first-boot cache test.
+

docs/v2/implementation/phase-6-2-authorization-runtime.md

@@ -0,0 +1,147 @@
+# Phase 6.2 — Authorization Runtime (ACL + LDAP grants)
+
+> **Status**: DRAFT — the v2 `plan.md` decision #129 + `acl-design.md` specify a 6-level permission-trie evaluator with `NodePermissions` bitmask grants, but no runtime evaluator exists. ACL tables are schematized but unread by the data path.
+>
+> **Branch**: `v2/phase-6-2-authorization-runtime`
+> **Estimated duration**: 2.5 weeks
+> **Predecessor**: Phase 6.1 (Resilience & Observability) — reuses the Polly pipeline for ACL-cache refresh retries
+> **Successor**: Phase 6.3 (Redundancy)
+
+## Phase Objective
+
+Wire ACL enforcement on every OPC UA Read / Write / Subscribe / Call path + LDAP group → admin role grants that the v2 plan specified but never ran. End-state: a user's effective permissions resolve through a per-session permission-trie over the 6-level `Cluster / Namespace / UnsArea / UnsLine / Equipment / Tag` hierarchy, cached per session, invalidated on generation-apply + LDAP group expiry.
+
+Closes these gaps:
+
+1. **Data-path ACL enforcement** — `NodeAcl` table + `NodePermissions` flags shipped; `NodeAclService.cs` present as a CRUD surface; no code consults ACLs at `Read`/`Write` time. OPC UA server answers everything to everyone.
+2. **`LdapGroupRoleMapping` for cluster-scoped admin grants** — decision #105 shipped as the *design*; admin roles are hardcoded (`FleetAdmin` / `ConfigEditor` / `ReadOnly`) with no cluster-scoping and no LDAP-to-grant table. Decision #105 explicitly lifts this from v2.1 into v2.0.
+3. **Explicit Deny pathway** — deferred to v2.1 (decision #129 note). Phase 6.2 ships *grants only*; `Deny` stays out.
+4. **Admin UI ACL grant editor** — `AclsTab.razor` exists but edits the now-unused `NodeAcl` table; needs to wire to the runtime evaluator + the new `LdapGroupRoleMapping` table.
+
+## Scope — What Changes
+
+**Architectural separation** (critical for correctness): `LdapGroupRoleMapping` is **control-plane only** — it maps LDAP groups to Admin UI roles (`FleetAdmin` / `ConfigEditor` / `ReadOnly`) and cluster scopes for Admin access. **It is NOT consulted by the OPC UA data-path evaluator.** The data-path evaluator reads `NodeAcl` rows joined directly against the session's **resolved LDAP group memberships**. The two concerns share zero runtime code path.
+
+| Concern | Change |
+|---------|--------|
+| `Configuration` project | New entity `LdapGroupRoleMapping { Id, LdapGroup, Role, ClusterId? (nullable = system-wide), IsSystemWide, GeneratedAtUtc }`. **Consumed only by Admin UI role routing.** Migration. Admin CRUD. |
+| `Core` → new `Core.Authorization` sub-namespace | `IPermissionEvaluator.Authorize(IEnumerable<Claim> identity, OpcUaOperation op, NodeId nodeId) → AuthorizationDecision`. `op` covers every OPC UA surface: Browse, Read, Write, HistoryRead, HistoryUpdate, CreateMonitoredItems, TransferSubscriptions, Call, Acknowledge, Confirm, Shelve. Result is tri-state (internal model distinguishes `Allow` / `NotGranted` / `Denied` + carries matched-grant provenance). Phase 6.2 only produces `Allow` + `NotGranted`; v2.1 Deny lands without API break. |
+| `PermissionTrieBuilder` | Builds trie from `NodeAcl` rows joined against **resolved LDAP group memberships**, keyed on 6-level scope hierarchy for Equipment namespaces. **SystemPlatform namespaces (Galaxy)** use a `FolderSegment` scope level between Namespace and Tag, populated from `Tag.FolderPath` segments, so folder subtree authorization works on Galaxy trees the same way UNS works on Equipment trees. Trie node carries `ScopeKind` enum. |
+| `PermissionTrieCache` + freshness | One trie per `(ClusterId, GenerationId)`. Invalidated on `sp_PublishGeneration` via in-process event bus AND generation-ID check on hot path — every authz call looks up `CurrentGenerationId` (Polly-wrapped, sub-second cache); a Backup that cached a stale generation detects the mismatch + forces re-load. **Redundancy-safe**. |
+| `UserAuthorizationState` freshness | Cached per session BUT bounded by `MembershipFreshnessInterval` (default **15 min**). Past that, the next hot-path authz call re-resolves LDAP group memberships via `LdapGroupService`. Failure to re-resolve (LDAP unreachable) → **fail-closed**: evaluator returns `NotGranted` for every call until memberships refresh successfully. Decoupled from Phase 6.1's availability-oriented 24h cache. |
+| `AuthCacheMaxStaleness` | Separate from Phase 6.1's `UsingStaleConfig` window. Default 5 min — beyond that, authz fails closed regardless of Phase 6.1 cache warmth. |
+| OPC UA server dispatch — all enforcement surfaces | `DriverNodeManager` wires evaluator on: **Browse + TranslateBrowsePathsToNodeIds** (ancestors implicitly visible if any descendant has a grant; denied ancestors filter from results), **Read** (per-attribute StatusCode `BadUserAccessDenied` in mixed-authorization batches; batch never poisons), **Write** (uses `NodePermissions.WriteOperate/Tune/Configure` based on driver `SecurityClassification`), **HistoryRead** (uses `NodePermissions.HistoryRead` — **distinct** flag, not Read), **HistoryUpdate** (`NodePermissions.HistoryUpdate`), **CreateMonitoredItems** (per-`MonitoredItemCreateResult` denial), **TransferSubscriptions** (re-evaluates items on transfer), **Call** (`NodePermissions.MethodCall`), **Acknowledge/Confirm/Shelve** (per-alarm flags). |
+| Subscription re-authorization | Each `MonitoredItem` is stamped with `(AuthGenerationId, MembershipVersion)` at create time. On every Publish, items with a stamp mismatching the session's current `(AuthGenerationId, MembershipVersion)` get re-evaluated; revoked items drop to `BadUserAccessDenied` within one publish cycle. Unchanged items stay fast-path. |
+| `LdapAuthService` | On cookie-auth success: resolves LDAP group memberships; loads matching `LdapGroupRoleMapping` rows → role claims + cluster-scope claims (control plane); stores `UserAuthorizationState.LdapGroups` on the session for the data-plane evaluator. |
+| `ValidatedNodeAclAuthoringService` | Replaces CRUD-only `NodeAclService` for authoring. Validates (LDAP group exists, scope exists in current or target draft, grant shape is valid, no duplicate `(LdapGroup, Scope)` pair). Admin UI writes only through it. |
+| Admin UI `AclsTab.razor` | Writes via `ValidatedNodeAclAuthoringService`. Adds Probe-This-Permission row that runs the real evaluator against a chosen `(LDAP group, node, operation)` and shows `Allow` / `NotGranted` + matched-grant provenance. |
+| Admin UI new tab `RoleGrantsTab.razor` | CRUD over `LdapGroupRoleMapping`. Per-cluster + system-wide grants. FleetAdmin only. **Documentation explicit** that this only affects Admin UI access, not OPC UA data plane. |
+| Audit log | Every Grant/Revoke/Publish on `LdapGroupRoleMapping` or `NodeAcl` writes an `AuditLog` row with old/new state + user. |
+
+## Scope — What Does NOT Change
+
+| Item | Reason |
+|------|--------|
+| OPC UA authn | Already done (PR 19 LDAP user identity + Basic256Sha256 profile). Phase 6.2 is authorization only. |
+| Explicit `Deny` grants | Decision #129 note explicitly defers to v2.1. Default-deny + additive grants only. |
+| Driver-side `SecurityClassification` metadata | Drivers keep reporting `Operate` / `ViewOnly` / etc. — the evaluator uses them as *part* of the decision but doesn't replace them. |
+| Galaxy namespace (SystemPlatform kind) | UNS levels don't apply; evaluator treats Galaxy nodes as `Cluster → Namespace → Tag` (skip UnsArea/UnsLine/Equipment). |
+
+## Entry Gate Checklist
+
+- [ ] Phase 6.1 merged (reuse `Core.Resilience` Polly pipeline for the ACL cache-refresh retries)
+- [ ] `acl-design.md` re-read in full
+- [ ] Decision log #105, #129, corrections-doc B1 re-skimmed
+- [ ] Existing `NodeAcl` + `NodePermissions` flag enum audited; confirm bitmask flags match `acl-design.md` table
+- [ ] Existing `LdapAuthService` group-resolution code path traced end-to-end — confirm it already queries group memberships (we only need the caller to consume the result)
+- [ ] Test DB scenarios catalogued: two clusters, three LDAP groups per cluster, mixed grant shapes; captured as seed-data fixtures
+
+## Task Breakdown
+
+### Stream A — `LdapGroupRoleMapping` table + migration (3 days)
+
+1. **A.1** Entity + EF Core migration. Columns per §Scope table. Unique constraint on `(LdapGroup, ClusterId)` with null-tolerant comparer for the system-wide case. Index on `LdapGroup` for the hot-path lookup on auth.
+2. **A.2** `ILdapGroupRoleMappingService` CRUD. Wrap in the Phase 6.1 Polly pipeline (timeout → retry → fallback-to-cache).
+3. **A.3** Seed-data migration: preserve the current hardcoded `FleetAdmin` / `ConfigEditor` / `ReadOnly` mappings by seeding rows for the existing LDAP groups the dev box uses (`cn=fleet-admin,…`, `cn=config-editor,…`, `cn=read-only,…`). Op no-op migration for existing deployments.
+
+### Stream B — Permission-trie evaluator (1 week)
+
+1. **B.1** `IPermissionEvaluator.Authorize(IEnumerable<Claim> identity, NodeId nodeId, NodePermissions needed)` — returns `bool`. Phase 6.2 returns only `true` / `false`; v2.1 can widen to `Allow`/`Deny`/`Indeterminate` if Deny lands.
+2. **B.2** `PermissionTrieBuilder` builds the trie from `NodeAcl` + `LdapGroupRoleMapping` joined to the current generation's `UnsArea` + `UnsLine` + `Equipment` + `Tag` tables. One trie per `(ClusterId, GenerationId)` so rollback doesn't smear permissions across generations.
+3. **B.3** Trie node structure: `{ Level: enum, ScopeId: Guid, AllowedPermissions: NodePermissions, ChildrenByLevel: Dictionary<Guid, TrieNode> }`. Evaluation walks from Cluster → Namespace → UnsArea → UnsLine → Equipment → Tag, ORing allowed permissions at each level. Additive semantics: a grant at Cluster level cascades to every descendant tag.
+4. **B.4** `PermissionTrieCache` service scoped as singleton; exposes `GetTrieAsync(ClusterId, ct)` that returns the current-generation trie. Invalidated on `sp_PublishGeneration` via an in-process event bus; also on TTL expiry (24 h safety net).
+5. **B.5** Per-session cached evaluator: OPC UA Session authentication produces `UserAuthorizationState { ClusterId, LdapGroups[], Trie }`; cached on the session until session close or generation-apply.
+6. **B.6** Unit tests: trie-walk theory covering (a) Cluster-level grant cascades to tags, (b) Equipment-level grant doesn't leak to sibling Equipment, (c) multi-group union, (d) no-grant → deny, (e) Galaxy nodes skip UnsArea/UnsLine levels.
+
+### Stream C — OPC UA server dispatch wiring (6 days, widened)
+
+1. **C.1** `DriverNodeManager.Read` — evaluator consulted per `ReadValueId` with `OpcUaOperation.Read`. Denied attributes get `BadUserAccessDenied` per-item; batch never poisons. Integration test covers mixed-authorization batch (3 authorized + 2 denied → 3 Good values + 2 Bad StatusCodes, request completes).
+2. **C.2** `DriverNodeManager.Write` — evaluator chooses `NodePermissions.WriteOperate` / `WriteTune` / `WriteConfigure` based on the driver-reported `SecurityClassification`.
+3. **C.3** `DriverNodeManager.HistoryRead` — **uses `NodePermissions.HistoryRead`**, which is a **distinct flag** from Read. Test: user with Read but not HistoryRead can read live values but gets `BadUserAccessDenied` on `HistoryRead`.
+4. **C.4** `DriverNodeManager.HistoryUpdate` — uses `NodePermissions.HistoryUpdate`.
+5. **C.5** `DriverNodeManager.CreateMonitoredItems` — per-`MonitoredItemCreateResult` denial in mixed-authorization batch; partial success path per OPC UA Part 4. Each created item stamped `(AuthGenerationId, MembershipVersion)`.
+6. **C.6** `DriverNodeManager.TransferSubscriptions` — on reconnect, re-evaluate every transferred `MonitoredItem` against the session's current auth state. Stale-stamp items drop to `BadUserAccessDenied`.
+7. **C.7** **Browse + TranslateBrowsePathsToNodeIds** — evaluator called with `OpcUaOperation.Browse`. Ancestor visibility implied when any descendant has a grant (per `acl-design.md` §Browse). Denied ancestors filter from browse results — the UA browser sees a hierarchy truncated at the denied ancestor rather than an inconsistent child-without-parent view.
+8. **C.8** `DriverNodeManager.Call` — `NodePermissions.MethodCall`.
+9. **C.9** Alarm actions (Acknowledge / Confirm / Shelve) — per-alarm `NodePermissions.AlarmAck` / `AlarmConfirm` / `AlarmShelve`.
+10. **C.10** Publish path — for each `MonitoredItem` with a mismatched `(AuthGenerationId, MembershipVersion)` stamp, re-evaluate. Unchanged items stay fast-path; changes happen at next publish cycle.
+11. **C.11** Integration tests: three-user seed with different memberships; matrix covers every operation in §Scope. Mixed-batch tests for Read + CreateMonitoredItems.
+
+### Stream D — Admin UI refresh (4 days)
+
+1. **D.1** `RoleGrantsTab.razor` — FleetAdmin-gated CRUD on `LdapGroupRoleMapping`. Per-cluster dropdown + system-wide checkbox. Validation: LDAP group must exist in the dev LDAP (GLAuth) before saving — best-effort probe with graceful degradation.
+2. **D.2** `AclsTab.razor` rewrites its edit path to write through the new `NodeAclService`. Adds a "Probe this permission" row: choose `(LDAP group, node, action)` → shows Allow / Deny + the reason (which grant matched).
+3. **D.3** Draft-generation diff viewer now includes an ACL section: "X grants added, Y grants removed, Z grants changed."
+4. **D.4** SignalR notification: `PermissionTrieCache` invalidation on `sp_PublishGeneration` pushes to Admin UI so operators see "this clusters permissions were just updated" within 2 s.
+
+## Compliance Checks (run at exit gate)
+
+- [ ] **Control/data-plane separation**: `LdapGroupRoleMapping` consumed only by Admin UI; the data-path evaluator has zero references to it. Enforced via a project-reference audit (Admin project references the mapping service; `Core.Authorization` does not).
+- [ ] **Every operation wired**: Browse, Read, Write, HistoryRead, HistoryUpdate, CreateMonitoredItems, TransferSubscriptions, Call, Acknowledge, Confirm, Shelve all consult the evaluator. Integration test matrix covers every operation × allow/deny.
+- [ ] **HistoryRead uses its own flag**: test "user with Read + no HistoryRead gets `BadUserAccessDenied` on HistoryRead".
+- [ ] **Mixed-batch semantics**: Read of 5 nodes (3 allowed + 2 denied) returns 3 Good + 2 `BadUserAccessDenied` per-`ReadValueId`; CreateMonitoredItems equivalent.
+- [ ] **Browse ancestor visibility**: user with a grant only on a deep equipment node can browse the path to it (ancestors implied); denied ancestors filter from browse results otherwise.
+- [ ] **Galaxy FolderSegment coverage**: a grant on a Galaxy folder subtree cascades to its tags; sibling folders are unaffected. Trie test covers this.
+- [ ] **Subscription re-authorization**: integration test — create item, revoke grant via draft+publish, next publish cycle the item returns `BadUserAccessDenied` (not silently still-notifying).
+- [ ] **Membership freshness**: test — 15 min MembershipFreshnessInterval elapses on a long-lived session + LDAP now unreachable → authz fails closed on the next request until LDAP recovers.
+- [ ] **Auth cache fail-closed**: test — Phase 6.1 cache serves stale config for 6 min; authz evaluator refuses all calls after 5 min regardless.
+- [ ] **Trie invariants**: `PermissionTrieBuilder` is idempotent (build twice with identical inputs → equal tries).
+- [ ] **Additive grants + cluster isolation**: cluster-grant cascades; cross-cluster leakage impossible.
+- [ ] **Redundancy-safe invalidation**: integration test — two nodes, a publish on one, authorize a request on the other before in-process event propagates → generation-mismatch forces re-load, no stale decision.
+- [ ] **Authoring validation**: `AclsTab` cannot save a `(LdapGroup, Scope)` pair that already exists in the draft; operator sees the validation error pre-save.
+- [ ] **AuthorizationDecision shape stability**: API surface exposes `Allow` + `NotGranted` only; `Denied` variant exists in the type but is never produced; v2.1 can add Deny without API break.
+- [ ] No regression in driver test counts.
+
+## Risks and Mitigations
+
+| Risk | Likelihood | Impact | Mitigation |
+|------|:----------:|:------:|------------|
+| ACL evaluator latency on per-read hot path | Medium | High | Trie lookup is O(depth) = O(6); session-cached UserAuthorizationState avoids per-Read trie rebuild; benchmark in Stream B.6 |
+| Trie cache stale after a rollback | Medium | High | `sp_PublishGeneration` + `sp_RollbackGeneration` both emit the invalidation event; trie keyed on `(ClusterId, GenerationId)` so rollback fetches the prior trie cleanly |
+| `BadUserAccessDenied` returns expose sensitive browse-name metadata | Low | Medium | Server returns only the status code + NodeId; no message leak per OPC UA Part 4 §7.34 guidance |
+| LdapGroupRoleMapping migration breaks existing deployments | Low | High | Seed-migration preserves the hardcoded groups' effective grants verbatim; smoke test exercises the post-migration fleet admin login |
+| Deny semantics accidentally ship (would break `acl-design.md` defer) | Low | Medium | `IPermissionEvaluator.Authorize` returns `bool` (not tri-state) through Phase 6.2; widening to `Allow`/`Deny`/`Indeterminate` is a v2.1 ticket |
+
+## Completion Checklist
+
+- [ ] Stream A: `LdapGroupRoleMapping` entity + migration + CRUD + seed
+- [ ] Stream B: evaluator + trie builder + cache + per-session state + unit tests
+- [ ] Stream C: OPC UA dispatch wiring on Read/Write/HistoryRead/Subscribe/Alarm paths
+- [ ] Stream D: Admin UI `RoleGrantsTab` + `AclsTab` refresh + SignalR invalidation
+- [ ] `phase-6-2-compliance.ps1` exits 0; exit-gate doc recorded
+
+## Adversarial Review — 2026-04-19 (Codex, thread `019da48d-0d2b-7171-aed2-fc05f1f39ca3`)
+
+1. **Crit · ACCEPT** — Trie must not conflate `LdapGroupRoleMapping` (control-plane admin claims per decision #105) with data-plane ACLs (decision #129). **Change**: `LdapGroupRoleMapping` is consumed only by the Admin UI role router. Data-plane trie reads `NodeAcl` rows joined against the session's **resolved LDAP groups**, never admin roles. Stream B.2 updated.
+2. **Crit · ACCEPT** — Cached `UserAuthorizationState` survives LDAP group changes because memberships only refresh at cookie-auth. Change: add `MembershipFreshnessInterval` (default 15 min); past that, next hot-path authz call forces group re-resolution (fail-closed if LDAP unreachable). Session-close-wins on config-rollback.
+3. **High · ACCEPT** — Node-local invalidation doesn't extend across redundant pair. **Change**: trie keyed on `(ClusterId, GenerationId)`; hot-path authz looks up `CurrentGenerationId` from the shared config DB (Polly-wrapped + sub-second cache). A Backup that read stale generation gets a mismatched trie → forces re-load. Implementation note added to Stream B.4.
+4. **High · ACCEPT** — Browse enforcement missing. **Change**: new Stream C.7 (`Browse + TranslateBrowsePathsToNodeIds` enforcement). Ancestor visibility implied when any descendant has a grant; denied ancestors filter from browse results per `acl-design.md` §Browse.
+5. **High · ACCEPT** — `HistoryRead` should use `NodePermissions.HistoryRead` bit, not `Read`. **Change**: Stream C.3 revised; separate unit test asserts `Read+no-HistoryRead` denies HistoryRead while allowing current-value reads.
+6. **High · ACCEPT** — Galaxy shallow-path (Cluster→Namespace→Tag) loses folder hierarchy authorization. **Change**: SystemPlatform namespaces use a `FolderSegment` scope-level between Namespace and Tag, populated from `Tag.FolderPath`; UNS-kind namespaces keep the 6-level hierarchy. Trie supports both via `ScopeKind` on each node.
+7. **High · ACCEPT** — Subscription re-authorization policy unresolved between create-time-only (fast, wrong on revoke) and per-publish (slow). **Change**: stamp each `MonitoredItem` with `(AuthGenerationId, MembershipVersion)`; re-evaluate on Publish only when either version changed. Revoked items drop to `BadUserAccessDenied` within one publish cycle.
+8. **Med · ACCEPT** — Mixed-authorization batch `Read` / `CreateMonitoredItems` service-result semantics underspecified. **Change**: Stream C.6 explicitly tests per-`ReadValueId` + per-`MonitoredItemCreateResult` denial in mixed batches; batch never collapses to a coarse failure.
+9. **Med · ACCEPT** — Missing surfaces: `Method.Call`, `HistoryUpdate`, event filter on subscriptions, subscription-transfer on reconnect, alarm-ack. **Change**: scope expanded — every OPC UA authorization surface enumerated in Stream C: Read, Write, HistoryRead, HistoryUpdate, CreateMonitoredItems, TransferSubscriptions, Call, Acknowledge/Confirm/Shelve, Browse, TranslateBrowsePathsToNodeIds.
+10. **Med · ACCEPT** — `bool` evaluator bakes in grant-only semantics; collides with v2.1 Deny. **Change**: internal model uses `AuthorizationDecision { Allow | NotGranted | Denied, IReadOnlyList<MatchedGrant> Provenance }`. Phase 6.2 maps `Denied` → never produced; UI + audit log use the full record so v2.1 Deny lands without API break.
+11. **Med · ACCEPT** — 6.1 cache fallback is availability-oriented; applying it to auth is correctness-dangerous. **Change**: auth-specific staleness budget `AuthCacheMaxStaleness` (default 5 min, not 24 h). Past that, hot-path evaluator fails closed on cached reads; all authorization calls return `NotGranted` until fresh data lands. Documented in risks + compliance.
+12. **Low · ACCEPT** — Existing `NodeAclService` is raw CRUD. **Change**: new `ValidatedNodeAclAuthoringService` enforces scope-uniqueness + draft/publish invariants + rejects invalid (LDAP group, scope) pairs; Admin UI writes through it only. Stream D.2 adjusted.
+

docs/v2/implementation/phase-6-3-redundancy-runtime.md

@@ -0,0 +1,150 @@
+# Phase 6.3 — Redundancy Runtime
+
+> **Status**: DRAFT — `CLAUDE.md` + `docs/Redundancy.md` describe a non-transparent warm/hot redundancy model with unique ApplicationUris, `RedundancySupport` advertisement, `ServerUriArray`, and dynamic `ServiceLevel`. Entities (`ServerCluster`, `ClusterNode`, `RedundancyRole`, `RedundancyMode`) exist; the runtime behavior (actual `ServiceLevel` number computation, mid-apply dip, `ServerUriArray` broadcast) is not wired.
+>
+> **Branch**: `v2/phase-6-3-redundancy-runtime`
+> **Estimated duration**: 2 weeks
+> **Predecessor**: Phase 6.2 (Authorization) — reuses the Phase 6.1 health endpoints for cluster-peer probing
+> **Successor**: Phase 6.4 (Admin UI completion)
+
+## Phase Objective
+
+Land the non-transparent redundancy protocol end-to-end: two `OtOpcUa.Server` instances in a `ServerCluster` each expose a live `ServiceLevel` node whose value reflects that instance's suitability to serve traffic, advertise each other via `ServerUriArray`, and transition role (Primary ↔ Backup) based on health + operator intent.
+
+Closes these gaps:
+
+1. **Dynamic `ServiceLevel`** — OPC UA Part 5 §6.3.34 specifies a Byte (0..255) that clients poll to pick the healthiest server. Our server publishes it as a static value today.
+2. **`ServerUriArray` broadcast** — Part 4 specifies that every node in a redundant pair should advertise its peers' ApplicationUris. Currently advertises only its own.
+3. **Primary / Backup role coordination** — entities carry `RedundancyRole` but the runtime doesn't read it; no peer health probing; no role-transfer on primary failure.
+4. **Mid-apply dip** — decision-level expectation that a server mid-generation-apply should report a *lower* ServiceLevel so clients cut over to the peer during the apply window. Not implemented.
+
+## Scope — What Changes
+
+| Concern | Change |
+|---------|--------|
+| `OtOpcUa.Server` → new `Server.Redundancy` sub-namespace | `RedundancyCoordinator` singleton. Resolves the current node's `ClusterNode` row at startup, loads peers, runs **two-layer peer health probe**: (a) `/healthz` every 2 s as the fast-fail (inherits Phase 6.1 semantics — HTTP + DB/cache healthy); (b) `UaHealthProbe` every 10 s — opens a lightweight OPC UA client session to the peer + reads its `ServiceLevel` node + verifies endpoint serves data. Authority decisions use UaHealthProbe; `/healthz` is used only to avoid wasting UA probes when peer is obviously down. |
+| Publish-generation fencing | Topology + role decisions are stamped with a monotonic `ConfigGenerationId` from the shared config DB. Coordinator re-reads topology via CAS on `(ClusterId, ExpectedGeneration)` → new row; peers reject state propagated from a lower generation. Prevents split-publish races. |
+| `InvalidTopology` runtime state | If both nodes detect >1 Primary AFTER startup (config-DB drift during a publish), both self-demote to ServiceLevel 2 until convergence. Neither node serves authoritatively; clients pick the healthier alternative or reconnect later. |
+| OPC UA server root | `ServiceLevel` variable node becomes a `BaseDataVariable` whose value updates on `RedundancyCoordinator` state change. `ServerUriArray` array variable includes **self + peers** in stable deterministic ordering (decision per OPC UA Part 4 §6.6.2.2). `RedundancySupport` stays static (set from `RedundancyMode` at startup); `Transparent` mode validated pre-publish, not rejected at startup. |
+| `RedundancyCoordinator` computation | **8-state ServiceLevel matrix** — avoids OPC UA Part 5 §6.3.34 collision (`0=Maintenance`, `1=NoData`). Operator-declared maintenance only = **0**. Unreachable / Faulted = **1**. In-range operational states occupy **2..255**: Authoritative-Primary = **255**; Isolated-Primary (peer unreachable, self serving) = **230**; Primary-Mid-Apply = **200**; Recovering-Primary (post-fault, dwell not met) = **180**; Authoritative-Backup = **100**; Isolated-Backup (primary unreachable, "take over if asked") = **80**; Backup-Mid-Apply = **50**; Recovering-Backup = **30**; `InvalidTopology` (runtime detects >1 Primary) = **2** (detected-inconsistency band — below normal operation). Full matrix documented in `docs/Redundancy.md` update. |
+| Role transition | Split-brain avoidance: role is *declared* in the shared config DB (`ClusterNode.RedundancyRole`), not elected at runtime. An operator flips the row (or a failover script does). Coordinator only reads; never writes. |
+| `sp_PublishGeneration` hook | Uses named **apply leases** keyed to `(ConfigGenerationId, PublishRequestId)`. `await using var lease = coordinator.BeginApplyLease(...)`. Disposal on any exit path (success, exception, cancellation) decrements. Watchdog auto-closes any lease older than `ApplyMaxDuration` (default 10 min) → ServiceLevel can't stick at mid-apply. Pre-publish validator rejects unsupported `RedundancyMode` (e.g. `Transparent`) with a clear error so runtime never sees an invalid state. |
+| Admin UI `/cluster/{id}` page | New `RedundancyTab.razor` — shows current node's role + ServiceLevel + peer reachability. FleetAdmin can trigger a role-swap by editing `ClusterNode.RedundancyRole` + publishing a draft. |
+| Metrics | New OpenTelemetry metrics: `ot_opcua_service_level{cluster,node}`, `ot_opcua_peer_reachable{cluster,node,peer}`, `ot_opcua_apply_in_progress{cluster,node}`. Sink via Phase 6.1 observability layer. |
+
+## Scope — What Does NOT Change
+
+| Item | Reason |
+|------|--------|
+| OPC UA authn / authz | Phases 6.2 + prior. Redundancy is orthogonal. |
+| Driver layer | Drivers aren't redundancy-aware; they run on each node independently against the same equipment. The server layer handles the ServiceLevel story. |
+| Automatic failover / election | Explicitly out of scope. Non-transparent = client picks which server to use via ServiceLevel + ServerUriArray. We do NOT ship consensus, leader election, or automatic promotion. Operator-driven failover is the v2.0 model per decision #79–85. |
+| Transparent redundancy (`RedundancySupport=Transparent`) | Not supported. If the operator asks for it the server fails startup with a clear error. |
+| Historian redundancy | Galaxy Historian's own redundancy (two historians on two CPUs) is out of scope. The Galaxy driver talks to whichever historian is reachable from its node. |
+
+## Entry Gate Checklist
+
+- [ ] Phase 6.1 merged (uses `/healthz` for peer probing)
+- [ ] `CLAUDE.md` §Redundancy + `docs/Redundancy.md` re-read
+- [ ] Decisions #79–85 re-skimmed
+- [ ] `ServerCluster`/`ClusterNode`/`RedundancyRole`/`RedundancyMode` entities + existing migration reviewed
+- [ ] OPC UA Part 4 §Redundancy + Part 5 §6.3.34 (ServiceLevel) re-skimmed
+- [ ] Dev box has two OtOpcUa.Server instances configured against the same cluster — one designated Primary, one Backup — for integration testing
+
+## Task Breakdown
+
+### Stream A — Cluster topology loader (3 days)
+
+1. **A.1** `RedundancyCoordinator` startup path: reads `ClusterNode` row for the current node (identified by `appsettings.json` `Cluster:NodeId`), reads the cluster's peer list, validates invariants (no duplicate `ApplicationUri`, at most one `Primary` per cluster if `RedundancyMode.WarmActive`, at most two nodes total in v2.0 per decision #83).
+2. **A.2** Topology subscription — coordinator re-reads on `sp_PublishGeneration` confirmation so an operator role-swap takes effect after publish (no process restart needed).
+3. **A.3** Tests: two-node cluster seed, one-node cluster seed (degenerate), duplicate-uri rejection.
+
+### Stream B — Peer health probing + ServiceLevel computation (6 days, widened)
+
+1. **B.1** `PeerHttpProbeLoop` per peer at 2 s — calls peer's `/healthz`, 1 s timeout, exponential backoff on sustained failure. Used as fast-fail.
+2. **B.2** `PeerUaProbeLoop` per peer at 10 s — opens an OPC UA client session to the peer (reuses Phase 5 `Driver.OpcUaClient` stack), reads peer's `ServiceLevel` node + verifies endpoint serves data. Short-circuit: if HTTP probe is failing, skip UA probe (no wasted sessions).
+3. **B.3** `ServiceLevelCalculator.Compute(role, selfHealth, peerHttpHealthy, peerUaHealthy, applyInProgress, recoveryDwellMet, topologyValid) → byte`. 8-state matrix per §Scope. `topologyValid=false` forces InvalidTopology = 2 regardless of other inputs.
+4. **B.4** `RecoveryStateManager`: after a `Faulted → Healthy` transition, hold driver in `Recovering` band (180 Primary / 30 Backup) for `RecoveryDwellTime` (default 60 s) AND require one positive publish witness (successful `Read` on a reference node) before entering Authoritative band.
+5. **B.5** Calculator reacts to inputs via `IObserver` so changes immediately push to the OPC UA `ServiceLevel` node.
+6. **B.6** Tests: **64-case matrix** covering role × self-health × peer-http × peer-ua × apply × recovery × topology. Specific cases flagged: Primary-with-unreachable-peer-serves-at-230 (authority retained); Backup-with-unreachable-primary-escalates-to-80 (not auto-promote); InvalidTopology demotes both nodes; Recovering dwell + publish-witness blocks premature return to 255.
+
+### Stream C — OPC UA node wiring (3 days)
+
+1. **C.1** `ServiceLevel` variable node created under `ServerStatus` at server startup. Type `Byte`, AccessLevel = CurrentRead only. Subscribe to `ServiceLevelCalculator` observable; push updates via `DataChangeNotification`.
+2. **C.2** `ServerUriArray` variable node under `ServerCapabilities`. Array of `String`, **includes self + peers** with deterministic ordering (self first). Updates on topology change. Compliance test asserts local-plus-peer membership.
+3. **C.3** `RedundancySupport` variable — static at startup from `RedundancyMode`. Values: `None`, `Cold`, `Warm`, `WarmActive`, `Hot`. Unsupported values (`Transparent`, `HotAndMirrored`) are rejected **pre-publish** by validator — runtime never sees them.
+4. **C.4** Client.CLI cutover test: connect to primary, read `ServiceLevel` → 255; pause primary apply → 200; unreachable peer while apply in progress → 200 (apply dominates peer-unreachable per matrix); client sees peer via `ServerUriArray`; fail primary → client reconnects to peer at 80 (isolated-backup band).
+
+### Stream D — Apply-window integration (3 days)
+
+1. **D.1** `sp_PublishGeneration` caller wraps the apply in `await using var lease = coordinator.BeginApplyLease(generationId, publishRequestId)`. Lease keyed to `(ConfigGenerationId, PublishRequestId)` so concurrent publishes stay isolated. Disposal decrements on every exit path.
+2. **D.2** `ApplyLeaseWatchdog` auto-closes leases older than `ApplyMaxDuration` (default 10 min) so a crashed publisher can't pin the node at mid-apply.
+3. **D.3** Pre-publish validator in `sp_PublishGeneration` rejects unsupported `RedundancyMode` values (`Transparent`, `HotAndMirrored`) with a clear error message — runtime never sees an invalid mode.
+4. **D.4** Tests: (a) mid-apply client subscribes → sees ServiceLevel drop → sees restore; (b) lease leak via `ThreadAbort` / cancellation → watchdog closes; (c) publish rejected for `Transparent` → operator-actionable error.
+
+### Stream E — Admin UI + metrics (3 days)
+
+1. **E.1** `RedundancyTab.razor` under `/cluster/{id}/redundancy`. Shows each node's role, current ServiceLevel (with band label per 8-state matrix), peer reachability (HTTP + UA probe separately), last apply timestamp. Role-swap button posts a draft edit on `ClusterNode.RedundancyRole`; publish applies.
+2. **E.2** OpenTelemetry meter export: `ot_opcua_service_level{cluster,node}` gauge + `ot_opcua_peer_reachable{cluster,node,peer,kind=http|ua}` + `ot_opcua_apply_in_progress{cluster,node}` + `ot_opcua_topology_valid{cluster}`. Sink via Phase 6.1 observability.
+3. **E.3** SignalR push: `FleetStatusHub` broadcasts ServiceLevel changes so the Admin UI updates within ~1 s of the coordinator observing a peer flip.
+
+### Stream F — Client-interoperability matrix (3 days, new)
+
+1. **F.1** Validate ServiceLevel-driven cutover against **Ignition 8.1 + 8.3**, **Kepware KEPServerEX 6.x**, **Aveva OI Gateway 2020R2 + 2023R1**. For each: configure the client with both endpoints, verify it honors `ServiceLevel` + `ServerUriArray` during primary failover.
+2. **F.2** Clients that don't honour the standards (doc field — may include Kepware and OI Gateway per Codex review) get an explicit compatibility-matrix entry: "requires manual backup-endpoint config / vendor-specific redundancy primitives". Documented in `docs/Redundancy.md`.
+3. **F.3** Galaxy MXAccess failover test — boot Galaxy.Proxy on both nodes, kill Primary, assert Galaxy consumer reconnects to Backup within `(SessionTimeout + KeepAliveInterval × 3)`. Document required session-timeout config in `docs/Redundancy.md`.
+
+## Compliance Checks (run at exit gate)
+
+- [ ] **OPC UA band compliance**: `0=Maintenance` reserved, `1=NoData` reserved. Operational states in 2..255 per 8-state matrix.
+- [ ] **Authoritative-Primary** ServiceLevel = 255.
+- [ ] **Isolated-Primary** (peer unreachable, self serving) = 230 — Primary retains authority.
+- [ ] **Primary-Mid-Apply** = 200.
+- [ ] **Recovering-Primary** = 180 with dwell + publish witness enforced.
+- [ ] **Authoritative-Backup** = 100.
+- [ ] **Isolated-Backup** (primary unreachable) = 80 — does NOT auto-promote.
+- [ ] **InvalidTopology** = 2 — both nodes self-demote when >1 Primary detected runtime.
+- [ ] **ServerUriArray** returns self + peer URIs, self first.
+- [ ] **UaHealthProbe authority**: integration test — peer returns HTTP 200 but OPC UA endpoint unreachable → coordinator treats peer as UA-unhealthy; peer is not a valid authority source.
+- [ ] **Apply-lease disposal**: leases close on exception, cancellation, and watchdog timeout; ServiceLevel never sticks at mid-apply band.
+- [ ] **Transparent-mode rejection**: attempting to publish `RedundancyMode=Transparent` is blocked at `sp_PublishGeneration`; runtime never sees an invalid mode.
+- [ ] **Role transition via operator publish**: FleetAdmin swaps `RedundancyRole` in a draft, publishes; both nodes re-read topology on publish confirmation + flip ServiceLevel — no restart.
+- [ ] **Client.CLI cutover**: with primary halted, Client.CLI that was connected to primary sees primary drop + reconnects to backup via `ServerUriArray`.
+- [ ] **Client interoperability matrix** (Stream F): Ignition 8.1 + 8.3 honour ServiceLevel; Kepware + Aveva OI Gateway findings documented.
+- [ ] **Galaxy MXAccess failover**: end-to-end test — primary kill → Galaxy consumer reconnects to backup within session-timeout budget.
+- [ ] No regression in existing driver test suites; no regression in `/healthz` reachability under redundancy load.
+
+## Risks and Mitigations
+
+| Risk | Likelihood | Impact | Mitigation |
+|------|:----------:|:------:|------------|
+| Split-brain from operator race (both nodes marked Primary) | Low | High | Coordinator rejects startup if its cluster has >1 Primary row; logs + fails fast. Document as a publish-time validation in `sp_PublishGeneration`. |
+| ServiceLevel thrashing on flaky peer | Medium | Medium | 2 s probe interval + 3-sample smoothing window; only declares a peer unreachable after 3 consecutive failed probes |
+| Client ignores ServiceLevel and stays on broken primary | Medium | Medium | Documented in `docs/Redundancy.md` — non-transparent redundancy requires client cooperation; most SCADA clients (Ignition, Kepware, Aveva OI Gateway) honor it. Unit-test the advertised values; field behavior is client-responsibility |
+| Apply-window counter leaks on exception | Low | High | `BeginApplyWindow` returns `IDisposable`; `using` syntax enforces paired decrement; unit test for exception-in-apply path |
+| `HttpClient` probe leaks sockets | Low | Medium | Single shared `HttpClient` per coordinator (not per-probe); timeouts tight to avoid keeping connections open during peer downtime |
+
+## Completion Checklist
+
+- [ ] Stream A: topology loader + tests
+- [ ] Stream B: peer probe + ServiceLevel calculator + 32-case matrix tests
+- [ ] Stream C: ServiceLevel / ServerUriArray / RedundancySupport node wiring + Client.CLI smoke test
+- [ ] Stream D: apply-window integration + nested-apply counter
+- [ ] Stream E: Admin `RedundancyTab` + OpenTelemetry metrics + SignalR push
+- [ ] `phase-6-3-compliance.ps1` exits 0; exit-gate doc; `docs/Redundancy.md` updated with the ServiceLevel matrix
+
+## Adversarial Review — 2026-04-19 (Codex, thread `019da490-3fa0-7340-98b8-cceeca802550`)
+
+1. **Crit · ACCEPT** — No publish-generation fencing enables split-publish advertising both as authoritative. **Change**: coordinator CAS on a monotonic `ConfigGenerationId`; every topology decision is generation-stamped; peers reject state propagated from a lower generation.
+2. **Crit · ACCEPT** — `>1 Primary` at startup covered but runtime containment missing when invalid topology appears later (mid-apply race). **Change**: add runtime `InvalidTopology` state — both nodes self-demote to ServiceLevel 2 (the "detected inconsistency" band, below normal operation) until convergence.
+3. **High · ACCEPT** — `0 = Faulted` collides with OPC UA Part 5 §6.3.34 semantics where 0 means **Maintenance** and 1 means NoData. **Change**: reserve **0** for operator-declared maintenance-mode only; Faulted/unreachable uses **1** (NoData); in-range degraded states occupy 2..199.
+4. **High · ACCEPT** — Matrix collapses distinct operational states onto the same value. **Change**: matrix expanded to Authoritative-Primary=255, Isolated-Primary=230 (peer unreachable — still serving), Primary-Mid-Apply=200, Recovering-Primary=180, Authoritative-Backup=100, Isolated-Backup=80 (primary unreachable — "take over if asked"), Backup-Mid-Apply=50, Recovering-Backup=30.
+5. **High · ACCEPT** — `/healthz` from 6.1 is HTTP-healthy but doesn't guarantee OPC UA data plane. **Change**: add a redundancy-specific probe `UaHealthProbe` — issues a `ReadAsync(ServiceLevel)` against the peer's OPC UA endpoint via a lightweight client session. `/healthz` remains the fast-fail; the UA probe is the authority signal.
+6. **High · ACCEPT** — `ServerUriArray` must include self + peers, not peers only. **Change**: array contains `[self.ApplicationUri, peer.ApplicationUri]` in stable deterministic ordering; compliance test asserts local-plus-peer membership.
+7. **Med · ACCEPT** — No `Faulted → Recovering → Healthy` path. **Change**: add `Recovering` state with min dwell time (60 s default) + positive publish witness (one successful Read on a reference node) before returning to Healthy. Thrash-prevention.
+8. **Med · ACCEPT** — Topology change during in-flight probe undefined. **Change**: every probe task tagged with `ConfigGenerationId` at dispatch; obsolete results discarded; in-flight probes cancelled on topology reload.
+9. **Med · ACCEPT** — Apply-window counter race on exception/cancellation/async ownership. **Change**: apply-window is a named lease keyed to `(ConfigGenerationId, PublishRequestId)` with disposal enforced via `await using`; watchdog detects leased-but-abandoned and force-closes after `ApplyMaxDuration` (default 10 min).
+10. **High · ACCEPT** — Ignition + Kepware + Aveva OI Gateway `ServiceLevel` compliance is unverified. **Change**: risk elevated to High; add Stream F (new) — build an interop matrix: validate against Ignition 8.1/8.3, Kepware KEPServerEX 6.x, Aveva OI Gateway 2020R2 + 2023R1. Document per-client cutover behaviour. Field deployments get a documented compatibility table; clients that ignore ServiceLevel documented as requiring explicit backup-endpoint config.
+11. **Med · ACCEPT** — Galaxy MXAccess re-session on Primary death not in acceptance. **Change**: Stream F adds an end-to-end failover smoke test that boots Galaxy.Proxy on both nodes, kills Primary, asserts Galaxy consumer reconnects to Backup within `(SessionTimeout + KeepAliveInterval × 3)` budget. `docs/Redundancy.md` updated with required session timeouts.
+12. **Med · ACCEPT** — Transparent-mode startup rejection is outage-prone. **Change**: `sp_PublishGeneration` validates `RedundancyMode` pre-publish — unsupported values reject the publish attempt with a clear validation error; runtime never sees an unsupported mode. Last-good config stays active.
+

docs/v2/implementation/phase-6-4-admin-ui-completion.md

@@ -0,0 +1,134 @@
+# Phase 6.4 — Admin UI Completion
+
+> **Status**: DRAFT — Phase 1 Stream E shipped the Admin scaffold + core pages; several feature-completeness items from its completion checklist (`phase-1-configuration-and-admin-scaffold.md` §Stream E) never landed. This phase closes them.
+>
+> **Branch**: `v2/phase-6-4-admin-ui-completion`
+> **Estimated duration**: 2 weeks
+> **Predecessor**: Phase 6.3 (Redundancy runtime) — reuses the `/cluster/{id}` page layout for the new tabs
+> **Successor**: v2 release-readiness capstone (Task #121)
+
+## Phase Objective
+
+Close the Admin UI feature-completeness checklist that Phase 1 Stream E exit gate left open. Each item below is an existing `phase-1-configuration-and-admin-scaffold.md` completion-checklist entry that is currently unchecked.
+
+Gaps to close:
+
+1. **UNS Structure tab drag/move with impact preview** — decision #115 + `admin-ui.md` §"UNS". Current state: list-only render; no drag reorder; no "X lines / Y equipment impacted" preview.
+2. **Equipment CSV import + 5-identifier search** — decision #95 + #117. Current state: basic form; no CSV parser; search indexes only ZTag.
+3. **Draft-generation diff viewer** — enhance existing `DiffViewer.razor` to show generation-diff not just staged-edit diff; highlight ACL grant changes (lands after Phase 6.2).
+4. **`_base` equipment-class Identification fields exposure** — decision #138–139. Columns exist on `Equipment`; no Admin UI field group; no address-space exposure of the OPC 40010 sub-folder.
+
+## Scope — What Changes
+
+| Concern | Change |
+|---------|--------|
+| `Admin/Pages/UnsTab.razor` | Tree component with drag-drop using **`MudBlazor.TreeView` + `MudBlazor.DropTarget`** (existing transitive dep — no new third-party package). Native HTML5 DnD rejected because virtualization + DnD on 500+ nodes doesn't combine reliably. Each drag fires a "Compute Impact" call carrying a `DraftRevisionToken`; modal preview ("Moving Line 'Oven-2' from 'Packaging' to 'Assembly' will re-home 14 equipment + re-parent 237 tags"). **Confirm step re-checks the token** and rejects with a `409 Conflict / refresh-required` modal if the draft advanced between preview and commit. |
+| `Admin/Services/UnsImpactAnalyzer.cs` | New service. Given a move-operation (line move, area rename, line merge), computes cascade counts + `DraftRevisionToken` at preview time. Pure-function shape; testable in isolation. |
+| `Admin/Pages/EquipmentTab.razor` | Add CSV-import button → modal with file picker + dry-run preview. **Identifier search** uses the canonical decision #117 set: `ZTag / MachineCode / SAPID / EquipmentId / EquipmentUuid`. Typeahead probes each column with a ranking query (exact match score 100 → prefix 50 → opt-in LIKE 20; published > draft tie-break). Result row shows which field matched via trailing badge. |
+| `Admin/Services/EquipmentCsvImporter.cs` | New service. CSV header row must start with `# OtOpcUaCsv v1` (version marker — future shape changes bump the version). Columns: `ZTag, MachineCode, SAPID, EquipmentId, EquipmentUuid, Name, UnsAreaName, UnsLineName, Manufacturer, Model, SerialNumber, HardwareRevision, SoftwareRevision, YearOfConstruction, AssetLocation, ManufacturerUri, DeviceManualUri`. Parser rejects unknown columns + blank required fields + duplicate ZTags + missing UnsLines. |
+| **Staged-import table** `EquipmentImportBatch` | New entity `{ Id, CreatedAtUtc, CreatedBy, RowsStaged, RowsAccepted, RowsRejected, FinalisedAtUtc? }` + child `EquipmentImportRow` records. Import writes rows in chunks to the staging table (not to `Equipment`). `FinaliseImportBatch` is the atomic finalize step that applies all accepted rows to `Equipment` + `ExternalIdReservation` in one transaction — short + bounded regardless of input size. Rollback = drop the batch row; `Equipment` never partially mutates. |
+| `Admin/Pages/DraftEditor.razor` + `DiffViewer.razor` | Diff viewer refactored into a base component + section plugins: `StructuralDiffSection`, `EquipmentDiffSection`, `TagDiffSection`, `AclDiffSection` (Phase 6.2), `RedundancyDiffSection` (Phase 6.3), `IdentificationDiffSection`. Each section has a **1000-row hard cap**; over-cap renders an aggregate summary + "Load full diff" button streaming 500-row pages via SignalR. Subtree-rename diffs (decision #115 bulk restructure) surface as summary only by default. |
+| `Admin/Components/IdentificationFields.razor` | New component. Renders the OPC 40010 field set **per decision #139**: `Manufacturer, Model, SerialNumber, HardwareRevision, SoftwareRevision, YearOfConstruction, AssetLocation, ManufacturerUri, DeviceManualUri`. `ProductInstanceUri / DeviceRevision / MonthOfConstruction` dropped from this phase — they need a separate decision-log widening. |
+| `OtOpcUa.Server/OpcUa/DriverNodeManager` — Equipment folder build | When an `Equipment` row has non-null Identification fields, the server adds an `Identification` sub-folder under the Equipment node containing one variable per non-null field. **ACL binding**: the sub-folder + variables inherit the `Equipment` scope's grants from Phase 6.2's trie — no new scope level added. Documented in `acl-design.md` cross-reference update. |
+
+## Scope — What Does NOT Change
+
+| Item | Reason |
+|------|--------|
+| Admin UI visual language | Bootstrap 5 / cookie auth / sidebar layout unchanged — consistency with ScadaLink design reference. |
+| LDAP auth flow | Already shipped in Phase 1. Phase 6.4 is additive UI only. |
+| Core abstractions / driver layer | Admin UI changes don't touch drivers. |
+| Equipment-class *template schema validation* | Still deferred (decision #112 — schemas repo not landed). We expose the Identification fields but don't validate against a template hierarchy. |
+| Drag/move to *other clusters* | Out of scope — equipment is cluster-scoped per decision #82. Cross-cluster migration is a different workflow. |
+
+## Entry Gate Checklist
+
+- [ ] Phase 6.2 merged (ACL grants are part of the new diff viewer sections)
+- [ ] Phase 6.3 merged (redundancy-role changes are part of the diff viewer)
+- [ ] `phase-1-configuration-and-admin-scaffold.md` §Stream E completion checklist re-read — confirm these are the remaining items
+- [ ] `admin-ui.md` re-skimmed for screen layouts
+- [ ] Existing `EquipmentTab.razor` / `UnsTab.razor` / `DraftEditor.razor` diff'd against what ships today so the edits are additive not destructive
+- [ ] Dev Galaxy available for OPC 40010 exposure smoke testing
+
+## Task Breakdown
+
+### Stream A — UNS drag/reorder + impact preview (5 days)
+
+1. **A.1** 1000-node synthetic seed fixture. Drag-latency bench against `MudBlazor.TreeView` + `MudBlazor.DropTarget` — commit to the component if latency budget (100 ms drag-enter feedback) holds; fall back to flat-list reorder UI (Area/Line dropdowns) with loss of visual drag affordance otherwise.
+2. **A.2** `UnsImpactAnalyzer` service. Inputs: `(DraftGenerationId, MoveOperation, DraftRevisionToken)`. Outputs: `ImpactPreview { AffectedEquipmentCount, AffectedTagCount, CascadeWarnings[], DraftRevisionToken }`. Pure-function shape; testable in isolation.
+3. **A.3** Modal preview wired to `UnsImpactAnalyzer`. **Confirm** re-reads the current draft revision + compares against the preview's token; if the draft advanced (another operator saved a different edit), show a `409 Conflict / refresh-required` modal rather than silently overwriting.
+4. **A.4** Cross-cluster drop attempts: target disabled + toast "Equipment is cluster-scoped (decision #82). To move across clusters, use Export → Import on the Cluster detail page." Plus help link.
+5. **A.5** Playwright (or equivalent) smoke test: drag a line across areas, assert modal shows right counts, assert draft row reflects the move; concurrent-edit test runs two sessions + asserts the later Confirm hits the 409.
+
+### Stream B — Equipment CSV import + 5-identifier search (5 days)
+
+1. **B.1** `EquipmentCsvImporter`. Strict RFC 4180 parser (per decision #95). Header row validation: first line must match `# OtOpcUaCsv v1` — future versions fork parser versions. Required columns: `ZTag, MachineCode, SAPID, EquipmentId, EquipmentUuid, Name, UnsAreaName, UnsLineName`. Optional: `Manufacturer, Model, SerialNumber, HardwareRevision, SoftwareRevision, YearOfConstruction, AssetLocation, ManufacturerUri, DeviceManualUri`. Parser rejects unknown columns + blank required fields + duplicate ZTags.
+2. **B.2** `EquipmentImportBatch` + `EquipmentImportRow` staging tables (migration). Import writes preview rows to staging via chunked inserts; staging never blocks `Equipment` or `ExternalIdReservation`. Preview query reads staging + validates each row against the current `Equipment` state + `ExternalIdReservation` freshness.
+3. **B.3** `ImportPreview` UI — per-row accept/reject table. Reject reasons: "ZTag already exists in draft", "ExternalIdReservation conflict with Cluster X", "UnsLineName not found in draft UNS tree", etc. Operator reviews + clicks "Commit".
+4. **B.4** `FinaliseImportBatch` — atomic finalize. One EF transaction applies accepted rows to `Equipment` + `ExternalIdReservation`; duration bounded regardless of input size (the atomic step is a bulk-insert, not per-row row-by-row). Rollback = drop batch row via `DropImportBatch`; `Equipment` never partially mutates.
+5. **B.5** Five-identifier search. Rank SQL: exact match any identifier = score 100, prefix match = 50, LIKE-fuzzy (opt-in via `?fuzzy=true`) = 20; tie-break `published > draft` then `RowVersion DESC`. Typeahead shows which field matched via trailing badge.
+6. **B.6** Smoke tests: 100-row CSV with 10 conflicts (5 ZTag dupes, 3 reservation clashes, 2 missing UnsLines); 10k-row perf test asserting finalize txn < 30 s; concurrent import + external `ExternalIdReservation` insert test asserts retryable-conflict handling.
+
+### Stream C — Diff viewer enhancements (4 days)
+
+1. **C.1** Refactor `DiffViewer.razor` into a base component + section plugins. Plugins: `StructuralDiffSection` (UNS tree), `EquipmentDiffSection`, `TagDiffSection`, `AclDiffSection` (Phase 6.2), `RedundancyDiffSection` (Phase 6.3), `IdentificationDiffSection`.
+2. **C.2** Each section renders collapsed by default; counts + top-line summary always visible. **1000-row hard cap** per section — over-cap sections render aggregate summary (e.g. "237 equipment re-parented from Packaging to Assembly") with a "Load full diff" button that streams 500-row pages via SignalR.
+3. **C.3** Subtree-rename diffs (decision #115 bulk restructure) surface as summary only by default regardless of row count.
+4. **C.4** Tests: seed two generations with deliberate diffs; assert every section reports the right counts + top-line summary + hard-cap behavior.
+
+### Stream D — OPC 40010 Identification exposure (3 days)
+
+1. **D.1** `IdentificationFields.razor` component. Renders the **9 decision #139 fields**: `Manufacturer, Model, SerialNumber, HardwareRevision, SoftwareRevision, YearOfConstruction, AssetLocation, ManufacturerUri, DeviceManualUri`. Labelled inputs; nullable columns show empty input; required-field validation on commit only.
+2. **D.2** `DriverNodeManager` equipment-folder builder — after building the equipment node, inspect the 9 Identification columns; if any non-null, add an `Identification` sub-folder with variable-per-non-null-field. ACL binding: sub-folder + variables inherit the **same `ScopeId` as the Equipment node** (Phase 6.2's trie treats them as part of the Equipment scope — no new scope level).
+3. **D.3** Address-space smoke test via Client.CLI: browse an equipment node, assert `Identification` sub-folder present when columns are set, absent when all null, variables match the field values.
+4. **D.4** ACL integration test: a user with Equipment-level grant reads the `Identification` variables without needing a separate grant; a user without the Equipment grant gets `BadUserAccessDenied` on both the Equipment node + its Identification variables.
+
+## Compliance Checks (run at exit gate)
+
+- [ ] **UNS drag/move**: drag a line across areas; modal preview shows correct impacted-equipment + impacted-tag counts.
+- [ ] **Concurrent-edit safety**: two-session test — session B saves a draft edit after session A opened the preview; session A's Confirm returns `409 Conflict / refresh-required` instead of overwriting.
+- [ ] **Cross-cluster drop**: dropping equipment across cluster boundaries is disabled + shows actionable toast pointing to Export/Import workflow.
+- [ ] **1000-node tree**: drag operations on a 1000-node seed maintain < 100 ms drag-enter feedback.
+- [ ] **CSV header version**: file missing `# OtOpcUaCsv v1` first line is rejected pre-parse.
+- [ ] **CSV canonical identifier set**: columns match decision #117 (ZTag / MachineCode / SAPID / EquipmentId / EquipmentUuid); drift from the earlier draft surfaces as a test failure.
+- [ ] **Staged-import atomicity**: `FinaliseImportBatch` transaction bounded < 30 s for a 10k-row import; pre-finalize stagings visible only to the importing user; rollback via `DropImportBatch`.
+- [ ] **Concurrent import + external reservation**: concurrent test — third party inserts to `ExternalIdReservation` mid-finalize; finalize retries with conflict handling; no corruption.
+- [ ] **5-identifier search ranking**: exact matches outrank prefix matches; published outranks draft for equal scores.
+- [ ] **Diff viewer section caps**: 2000-row subtree-rename diff renders as summary only; "Load full diff" streams in pages.
+- [ ] **OPC 40010 field list match**: rendered field group matches decision #139 exactly; no extra fields.
+- [ ] **OPC 40010 exposure**: Client.CLI browse shows `Identification` sub-folder when equipment has non-null columns; absent when all null.
+- [ ] **ACL inheritance for Identification**: integration test — Equipment-grant user reads Identification; no-grant user gets `BadUserAccessDenied` on both.
+- [ ] **Visual parity reviewer**: named role (`FleetAdmin` user, not the implementation lead) compares side-by-side against `admin-ui.md` §Visual-Design reference panels; signoff artefact is a checked-in screenshot set under `docs/v2/visual-compliance/phase-6-4/`.
+
+## Risks and Mitigations
+
+| Risk | Likelihood | Impact | Mitigation |
+|------|:----------:|:------:|------------|
+| UNS drag-drop janky on large trees (>500 nodes) | Medium | Medium | Virtualize the tree component; default-collapse nested areas; test with a synthetic 1000-equipment seed |
+| CSV import performance on 10k-row imports | Medium | Medium | Stream-parse rather than load-into-memory; preview renders in batches of 100; commit is chunked-EF-insert with progress bar |
+| Diff viewer becomes unwieldy with many sections | Low | Medium | Each section collapsed by default; top-line summary row always shown; Phase 6.4 caps at 6 sections |
+| OPC 40010 sub-folder accidentally exposes NULL/empty identification columns as empty-string variables | Low | Low | Column null-check in the builder; drop variables whose DB value is null |
+| 5-identifier search pulls full table | Medium | Medium | Indexes on each of ZTag/SAPID/UniqueId/Alias1/Alias2; search query uses a UNION of 5 indexed lookups; falls back to LIKE only on explicit operator opt-in |
+
+## Completion Checklist
+
+- [ ] Stream A: `UnsImpactAnalyzer` + drag-drop tree + modal preview + Playwright smoke
+- [ ] Stream B: `EquipmentCsvImporter` + preview modal + 5-identifier search + conflict-rollback test
+- [ ] Stream C: `DiffViewer` refactor + 6 section plugins + 2-generation diff test
+- [ ] Stream D: `IdentificationFields.razor` + address-space builder change + Client.CLI browse test
+- [ ] Visual-compliance reviewer signoff
+- [ ] Full solution `dotnet test` passes; `phase-6-4-compliance.ps1` exits 0; exit-gate doc
+
+## Adversarial Review — 2026-04-19 (Codex, via `codex-rescue` subagent)
+
+1. **Crit · ACCEPT** — Stale UNS impact preview can overwrite concurrent draft edits. **Change**: each preview carries a `DraftRevisionToken`; `Confirm` compares against the current draft + rejects with a `409 Conflict / refresh-required` modal if any draft edit landed since the preview was generated. Stream A.3 updated.
+2. **High · ACCEPT** — CSV import atomicity is internally contradictory (single EF transaction vs. chunked inserts). **Change**: one explicit model — staged-import table (`EquipmentImportBatch { Id, CreatedAtUtc, RowsStaged, RowsAccepted, RowsRejected }`) receives rows in chunks; final `FinaliseImportBatch` is atomic over `Equipment` + `ExternalIdReservation`. Rollback is "drop the batch row" — the real Equipment table is never partially mutated.
+3. **Crit · ACCEPT** — Identifier contract rewrite mis-cites decisions. **Change**: revert to the `admin-ui.md` + decision #117 canonical set — `ZTag / MachineCode / SAPID / EquipmentId / EquipmentUuid`. CSV header follows that set verbatim. Introduce a separate decision entry for versioned CSV header shape before adding any new column; CSV header row must start with `# OtOpcUaCsv v1` so future shape changes are unambiguous.
+4. **Med · ACCEPT** — Search ordering undefined. **Change**: rank SQL — exact match on any identifier scores 100; prefix match 50; LIKE-fuzzy 20; published > draft tie-breaker; `ORDER BY score DESC, RowVersion DESC`. Typeahead shows which field matched via trailing badge.
+5. **High · ACCEPT** — HTML5 DnD on virtualized tree is aspirational. **Change**: Stream A.2 rewritten — commits to **`MudBlazor.TreeView` + `MudBlazor.DropTarget`** (already a transitive dep via the existing Admin UI). Build a 1000-node synthetic seed in A.1 + validate drag-latency budget before implementing impact preview. If MudBlazor can't hit the budget, fall back to a flat-list reorder UI with Area/Line dropdowns (loss of visual drag affordance but unblocks the feature).
+6. **Med · ACCEPT** — Collapsed-by-default doesn't handle generation-sized diffs. **Change**: each diff section has a hard row cap (1000 by default). Over-cap sections render an aggregate summary + "Load full diff" button that streams via SignalR in 500-row pages. Decision #115 subtree renames surface as a "N equipment re-parented under X → Y" summary instead of row-by-row.
+7. **High · ACCEPT** — OPC 40010 field list doesn't match decision #139. **Change**: field group realigned to `Manufacturer, Model, SerialNumber, HardwareRevision, SoftwareRevision, YearOfConstruction, AssetLocation, ManufacturerUri, DeviceManualUri`. `ProductInstanceUri / DeviceRevision / MonthOfConstruction` dropped from Phase 6.4 — they belong to a future OPC 40010 widening decision.
+8. **High · ACCEPT** — `Identification` subtree unreconciled with ACL hierarchy (Phase 6.2 6-level scope). **Change**: address-space builder creates the Identification sub-folder under the Equipment node **with the same ScopeId as Equipment** — no new scope level. ACL evaluator treats `…/Equipment/Identification/X` as inheriting the `Equipment` scope's grants. Documented in Phase 6.2's `acl-design.md` cross-reference update.
+9. **Low · ACCEPT** — Visual-review gate names nonexistent reviewer role. **Change**: rubric defined — a named "Admin UX reviewer" (role `FleetAdmin` user, not the implementation lead) compares side-by-side screenshots against the `admin-ui.md` §Visual-Design reference panels; signoff artefact is a checked-in screenshot set under `docs/v2/visual-compliance/phase-6-4/`.
+10. **Med · ACCEPT** — Cross-cluster drag/drop lacks loud failure path. **Change**: on drop across cluster boundary, disable the drop target + show a toast "Equipment is cluster-scoped (decision #82). To move across clusters, use the Export → Import workflow on the Cluster detail page." Plus a help link. Tested in Stream A.4.
+

docs/v2/plan.md

@@ -909,6 +909,26 @@ Each step leaves the system runnable. The generic extraction is effectively free
 | 140 | Enterprise shortname = `zb` (UNS level-1 segment) | Closes corrections-doc D4. Matches the existing `ZB.MOM.WW.*` namespace prefix used throughout the codebase; short by design since this segment appears in every equipment path (`zb/warsaw-west/bldg-3/line-2/cnc-mill-05/RunState`); operators already say "ZB" colloquially. Admin UI cluster-create form default-prefills `zb` for the Enterprise field. Production deployments use it directly from cluster-create | 2026-04-17 |
 | 141 | Tier 3 (AppServer IO) cutover is feasible — AVEVA's OI Gateway supports arbitrary upstream OPC UA servers as a documented pattern | Closes corrections-doc E2 with **GREEN-YELLOW** verdict. Multiple AVEVA partners (Software Toolbox, InSource) have published working integrations against four different non-AVEVA upstream servers (TOP Server, OPC Router, OmniServer, Cogent DataHub). No re-architecting of OtOpcUa required. Path: `OPC UA node → OI Gateway → SuiteLink → $DDESuiteLinkDIObject → AppServer attribute`. Recommended AppServer floor: System Platform 2023 R2 Patch 01. Two integrator-burden risks tracked: validation/GxP paperwork (no AVEVA blueprint exists for non-AVEVA upstream servers in Part 11 deployments) and unpublished scale benchmarks (in-house benchmark required before cutover scheduling). See `aveva-system-platform-io-research.md` | 2026-04-17 |
 | 142 | Phase 1 acceptance includes an end-to-end AppServer-via-OI-Gateway smoke test against OtOpcUa | Catches AppServer-specific quirks (cert exchange via reject-and-trust workflow, endpoint URL must NOT include `/discovery` suffix per Inductive Automation forum failure mode, service-account install required because OI Gateway under SYSTEM cannot connect to remote OPC servers, `Basic256Sha256` + `SignAndEncrypt` + LDAP-username token combination must work end-to-end) early — well before the Year 3 tier-3 cutover schedule. Adds one task to `phase-1-configuration-and-admin-scaffold.md` Stream E (Admin smoke test) | 2026-04-17 |
+| 143 | Polly per-capability policy — Read / HistoryRead / Discover / Probe / Alarm-subscribe auto-retry; Write does NOT auto-retry unless the tag metadata carries `[WriteIdempotent]` | Decisions #44-45 forbid auto-retry on Write because a timed-out write can succeed on the device + be replayed by the pipeline, duplicating pulses / alarm acks / counter increments / recipe-step advances. Per-capability policy in the shared Polly layer makes the retry safety story explicit; `WriteIdempotentAttribute` on tag definitions is the opt-in surface | 2026-04-19 |
+| 144 | Polly pipeline key = `(DriverInstanceId, HostName)`, not DriverInstanceId alone | Decision #35 requires per-device isolation. One dead PLC behind a multi-device Modbus driver must NOT open the circuit breaker for healthy sibling hosts. Per-instance pipelines would poison every device behind one bad endpoint | 2026-04-19 |
+| 145 | Tier A/B/C runtime enforcement splits into `MemoryTracking` (all tiers — soft/hard thresholds log + surface, NEVER kill) and `MemoryRecycle` (Tier C only — requires out-of-process topology). Tier A/B hard-breach logs a promotion-to-Tier-C recommendation; the runtime never auto-kills an in-process driver | Decisions #73-74 reserve process-kill protections for Tier C. An in-process Tier A/B "recycle" would kill every OPC UA session + every other in-proc driver for one leaky instance, blast-radius worse than the leak | 2026-04-19 |
+| 146 | Memory watchdog uses the hybrid formula `soft = max(multiplier × baseline, baseline + floor)`, with baseline captured as the median of the first 5 min of `GetMemoryFootprint()` samples post-InitializeAsync. Tier-specific constants: A multiplier=3 floor=50 MB, B multiplier=3 floor=100 MB, C multiplier=2 floor=500 MB. Hard = 2 × soft | Codex adversarial review on the Phase 6.1 plan flagged that hardcoded per-tier MB bands diverge from decision #70's specified formula. Static bands false-trigger on small-footprint drivers + miss meaningful growth on large ones. Observed-baseline + hybrid formula recovers the original intent | 2026-04-19 |
+| 147 | `WedgeDetector` uses demand-aware criteria `(state==Healthy AND hasPendingWork AND noProgressIn > threshold)`. `hasPendingWork` = (Polly bulkhead depth > 0) OR (active MonitoredItem count > 0) OR (queued historian read count > 0). Idle + subscription-only + write-only-burst drivers stay Healthy without false-fault | Previous "no successful Read in N intervals" formulation flipped legitimate idle subscribers, slow historian backfills, and write-heavy drivers to Faulted. The demand-aware check only fires when the driver claims work is outstanding | 2026-04-19 |
+| 148 | LiteDB config cache is **generation-sealed**: `sp_PublishGeneration` writes `<cache-root>/<cluster>/<generationId>.db` as a read-only sealed file; cache reads serve the last-known-sealed generation. Mixed-generation reads are impossible | Prior "refresh on every successful query" cache could serve LDAP role mapping from one generation alongside UNS topology from another, producing impossible states. Sealed-snapshot invariant keeps cache-served reads coherent with a real published state | 2026-04-19 |
+| 149 | `AuthorizationDecision { Allow \| NotGranted \| Denied, IReadOnlyList<MatchedGrant> Provenance }` — tri-state internal model. Phase 6.2 only produces `Allow` + `NotGranted` (grant-only semantics per decision #129); v2.1 Deny widens without API break | bool return would collapse `no-matching-grant` and `explicit-deny` into the same runtime state + UI explanation; provenance record is needed for the audit log anyway. Making the shape tri-state from Phase 6.2 avoids a breaking change in v2.1 | 2026-04-19 |
+| 150 | Data-plane ACL evaluator consumes `NodeAcl` rows joined against the session's resolved LDAP group memberships. `LdapGroupRoleMapping` (decision #105) is control-plane only — routes LDAP groups to Admin UI roles. Zero runtime overlap between the two | Codex adversarial review flagged that Phase 6.2 draft conflated the two — building the data-plane trie from `LdapGroupRoleMapping` would let a user inherit tag permissions from an admin-role claim path never intended as a data-path grant | 2026-04-19 |
+| 151 | `UserAuthorizationState` cached per session but bounded by `MembershipFreshnessInterval` (default 15 min). Past that interval the next hot-path authz call re-resolves LDAP group memberships; failure to re-resolve (LDAP unreachable) → fail-closed (evaluator returns `NotGranted` until memberships refresh successfully) | Previous design cached memberships until session close, so a user removed from a privileged LDAP group could keep authorized access for hours. Bounded freshness + fail-closed covers the revoke-takes-effect story | 2026-04-19 |
+| 152 | Auth cache has its own staleness budget `AuthCacheMaxStaleness` (default 5 min), independent of decision #36's availability-oriented config cache (24 h). Past 5 min on authorization data, evaluator fails closed regardless of whether the underlying config is still serving from cache | Availability-oriented caches trade correctness for uptime. Authorization data is correctness-sensitive — stale ACLs silently extend revoked access. Auth-specific budget keeps the two concerns from colliding | 2026-04-19 |
+| 153 | MonitoredItem carries `(AuthGenerationId, MembershipVersion)` stamp at create time. On every Publish, items with a mismatching stamp re-evaluate; unchanged items stay fast-path. Revoked items drop to `BadUserAccessDenied` within one publish cycle | Create-time-only authorization leaves revoked users receiving data forever; per-publish re-authorization at 100 ms cadence across 50 groups × 6 levels is too expensive. Stamp-then-reevaluate-on-change balances correctness with cost | 2026-04-19 |
+| 154 | ServiceLevel reserves `0` for operator-declared maintenance only; `1` = NoData (unreachable / Faulted); operational states occupy `2..255` in an 8-state matrix (Authoritative-Primary=255, Isolated-Primary=230, Primary-Mid-Apply=200, Recovering-Primary=180, Authoritative-Backup=100, Isolated-Backup=80, Backup-Mid-Apply=50, Recovering-Backup=30, InvalidTopology=2) | OPC UA Part 5 §6.3.34 defines `0=Maintenance` + `1=NoData`; using `0` for our Faulted case collides with spec + triggers spec-compliant clients to enter maintenance-mode cutover. Expanded 8-state matrix covers operational states the 5-state original collapsed together (e.g. Isolated-Primary vs Primary-Mid-Apply were both 200) | 2026-04-19 |
+| 155 | `ServerUriArray` includes self + peers (self first, deterministic ordering), per OPC UA Part 4 §6.6.2.2 | Previous design excluded self from the array — spec violation + clients lose the ability to map server identities consistently during failover | 2026-04-19 |
+| 156 | Redundancy peer health uses a two-layer probe: `/healthz` (2 s) as fast-fail + `UaHealthProbe` (10 s, opens OPC UA client session to peer + reads its `ServiceLevel` node) as the authority signal. HTTP-healthy ≠ UA-authoritative | `/healthz` returns 200 whenever HTTP + config DB/cache is healthy — but a peer can be HTTP-healthy with a broken OPC UA endpoint or a stuck subscription publisher. Using HTTP alone would advertise authority against servers that can't actually publish data | 2026-04-19 |
+| 157 | Publish-generation fencing — coordinator CAS on a monotonic `ConfigGenerationId`; every topology + role decision is generation-stamped; peers reject state propagated from a lower generation. Runtime `InvalidTopology` state (both self-demote to ServiceLevel 2) when >1 Primary detected post-startup | Operator race publishing two drafts with different roles can produce two locally-valid views; without fencing + runtime containment both nodes can serve as Primary until manual intervention | 2026-04-19 |
+| 158 | Apply-window uses named leases keyed to `(ConfigGenerationId, PublishRequestId)` via `await using`. `ApplyLeaseWatchdog` auto-closes any lease older than `ApplyMaxDuration` (default 10 min) | Simple `IDisposable`-counter design leaks on cancellation / async-ownership races; a stuck positive count leaves the node permanently mid-apply. Generation-keyed leases + watchdog bound worst case | 2026-04-19 |
+| 159 | CSV import header row must start with `# OtOpcUaCsv v1` (version marker). Future shape changes bump the version; parser forks per version. Canonical identifier columns follow decision #117: `ZTag, MachineCode, SAPID, EquipmentId, EquipmentUuid` | Without a version marker the CSV schema has no upgrade path — adding a required column breaks every old export silently. The version prefix makes parser dispatch explicit + future-compatible | 2026-04-19 |
+| 160 | Equipment CSV import uses a staged-import pattern: `EquipmentImportBatch` + `EquipmentImportRow` tables receive chunked inserts; `FinaliseImportBatch` is one atomic transaction that applies accepted rows to `Equipment` + `ExternalIdReservation`. Rollback = drop the batch row; `Equipment` never partially mutates | 10k-row single-transaction import holds locks too long; chunked direct writes lose all-or-nothing rollback. Staging + atomic finalize bounds transaction duration + preserves rollback semantics | 2026-04-19 |
+| 161 | UNS drag-reorder impact preview carries a `DraftRevisionToken`; Confirm re-checks against the current draft + returns `409 Conflict / refresh-required` if the draft advanced between preview and commit | Without concurrency control, two operators editing the same draft can overwrite each other's changes silently. Draft-revision token + 409 response makes the race visible + forces refresh | 2026-04-19 |
+| 162 | OPC 40010 Identification sub-folder exposed under each equipment node inherits the Equipment scope's ACL grants — the ACL trie does NOT add a new scope level for Identification | Adding a new scope level for Identification would require every grant to add a second grant for `Equipment/Identification`; inheriting the Equipment scope keeps the grant model flat + prevents operator-forgot-to-grant-Identification access surprises | 2026-04-19 |
 
 ## Reference Documents
 

scripts/compliance/phase-6-1-compliance.ps1

@@ -0,0 +1,79 @@
+<#
+.SYNOPSIS
+    Phase 6.1 exit-gate compliance check — stub. Each `Assert-*` either passes
+    (Write-Host green) or throws. Non-zero exit = fail.
+
+.DESCRIPTION
+    Validates Phase 6.1 (Resilience & Observability runtime) completion. Checks
+    enumerated in `docs/v2/implementation/phase-6-1-resilience-and-observability.md`
+    §"Compliance Checks (run at exit gate)".
+
+    Current status: SCAFFOLD. Every check writes a TODO line and does NOT throw.
+    Each implementation task in Phase 6.1 is responsible for replacing its TODO
+    with a real check before closing that task.
+
+.NOTES
+    Usage:  pwsh ./scripts/compliance/phase-6-1-compliance.ps1
+    Exit:   0 = all checks passed (or are still TODO); non-zero = explicit fail
+#>
+[CmdletBinding()]
+param()
+
+$ErrorActionPreference = 'Stop'
+$script:failures = 0
+
+function Assert-Todo {
+    param([string]$Check, [string]$ImplementationTask)
+    Write-Host "  [TODO] $Check  (implement during $ImplementationTask)" -ForegroundColor Yellow
+}
+
+function Assert-Pass {
+    param([string]$Check)
+    Write-Host "  [PASS] $Check" -ForegroundColor Green
+}
+
+function Assert-Fail {
+    param([string]$Check, [string]$Reason)
+    Write-Host "  [FAIL] $Check — $Reason" -ForegroundColor Red
+    $script:failures++
+}
+
+Write-Host ""
+Write-Host "=== Phase 6.1 compliance — Resilience & Observability runtime ===" -ForegroundColor Cyan
+Write-Host ""
+
+Write-Host "Stream A — Resilience layer"
+Assert-Todo "Invoker coverage — every capability-interface method routes through CapabilityInvoker (analyzer error-level)" "Stream A.3"
+Assert-Todo "Write-retry guard — writes without [WriteIdempotent] never retry" "Stream A.5"
+Assert-Todo "Pipeline isolation — `(DriverInstanceId, HostName)` key; one dead host does not open breaker for siblings" "Stream A.5"
+
+Write-Host ""
+Write-Host "Stream B — Tier A/B/C runtime"
+Assert-Todo "Tier registry — every driver type has non-null Tier; Tier C declares out-of-process topology" "Stream B.1"
+Assert-Todo "MemoryTracking never kills — soft/hard breach on Tier A/B logs + surfaces without terminating" "Stream B.6"
+Assert-Todo "MemoryRecycle Tier C only — hard breach on Tier A never invokes supervisor; Tier C does" "Stream B.6"
+Assert-Todo "Wedge demand-aware — idle/historic-backfill/write-only cases stay Healthy" "Stream B.6"
+Assert-Todo "Galaxy supervisor preserved — Driver.Galaxy.Proxy/Supervisor/CircuitBreaker + Backoff still present + invoked" "Stream A.4"
+
+Write-Host ""
+Write-Host "Stream C — Health + logging"
+Assert-Todo "Health state machine — /healthz + /readyz respond < 500 ms for every DriverState per matrix in plan" "Stream C.4"
+Assert-Todo "Structured log — CI grep asserts DriverInstanceId + CorrelationId JSON fields present" "Stream C.4"
+
+Write-Host ""
+Write-Host "Stream D — LiteDB cache"
+Assert-Todo "Generation-sealed snapshot — SQL kill mid-op serves last-sealed snapshot; UsingStaleConfig=true" "Stream D.4"
+Assert-Todo "Mixed-generation guard — corruption of snapshot file fails closed; no mixed reads" "Stream D.4"
+Assert-Todo "First-boot no-snapshot + DB-down — InitializeAsync fails with clear error" "Stream D.4"
+
+Write-Host ""
+Write-Host "Cross-cutting"
+Assert-Todo "No test-count regression — dotnet test ZB.MOM.WW.OtOpcUa.slnx count ≥ pre-Phase-6.1 baseline" "Final exit-gate"
+
+Write-Host ""
+if ($script:failures -eq 0) {
+    Write-Host "Phase 6.1 compliance: scaffold-mode PASS (all checks TODO)" -ForegroundColor Green
+    exit 0
+}
+Write-Host "Phase 6.1 compliance: $script:failures FAIL(s)" -ForegroundColor Red
+exit 1

scripts/compliance/phase-6-2-compliance.ps1

@@ -0,0 +1,81 @@
+<#
+.SYNOPSIS
+    Phase 6.2 exit-gate compliance check — stub. Each `Assert-*` either passes
+    (Write-Host green) or throws. Non-zero exit = fail.
+
+.DESCRIPTION
+    Validates Phase 6.2 (Authorization runtime) completion. Checks enumerated
+    in `docs/v2/implementation/phase-6-2-authorization-runtime.md`
+    §"Compliance Checks (run at exit gate)".
+
+    Current status: SCAFFOLD. Every check writes a TODO line and does NOT throw.
+    Each implementation task in Phase 6.2 is responsible for replacing its TODO
+    with a real check before closing that task.
+
+.NOTES
+    Usage:  pwsh ./scripts/compliance/phase-6-2-compliance.ps1
+    Exit:   0 = all checks passed (or are still TODO); non-zero = explicit fail
+#>
+[CmdletBinding()]
+param()
+
+$ErrorActionPreference = 'Stop'
+$script:failures = 0
+
+function Assert-Todo {
+    param([string]$Check, [string]$ImplementationTask)
+    Write-Host "  [TODO] $Check  (implement during $ImplementationTask)" -ForegroundColor Yellow
+}
+
+function Assert-Pass {
+    param([string]$Check)
+    Write-Host "  [PASS] $Check" -ForegroundColor Green
+}
+
+function Assert-Fail {
+    param([string]$Check, [string]$Reason)
+    Write-Host "  [FAIL] $Check — $Reason" -ForegroundColor Red
+    $script:failures++
+}
+
+Write-Host ""
+Write-Host "=== Phase 6.2 compliance — Authorization runtime ===" -ForegroundColor Cyan
+Write-Host ""
+
+Write-Host "Stream A — LdapGroupRoleMapping (control plane)"
+Assert-Todo "Control/data-plane separation — Core.Authorization has zero refs to LdapGroupRoleMapping" "Stream A.2"
+Assert-Todo "Authoring validation — AclsTab rejects duplicate (LdapGroup, Scope) pre-save" "Stream A.3"
+
+Write-Host ""
+Write-Host "Stream B — Evaluator + trie + cache"
+Assert-Todo "Trie invariants — PermissionTrieBuilder idempotent (build twice == equal)" "Stream B.1"
+Assert-Todo "Additive grants + cluster isolation — cross-cluster leakage impossible" "Stream B.1"
+Assert-Todo "Galaxy FolderSegment coverage — folder-subtree grant cascades; siblings unaffected" "Stream B.2"
+Assert-Todo "Redundancy-safe invalidation — generation-mismatch forces trie re-load on peer" "Stream B.4"
+Assert-Todo "Membership freshness — 15 min interval elapsed + LDAP down = fail-closed" "Stream B.5"
+Assert-Todo "Auth cache fail-closed — 5 min AuthCacheMaxStaleness exceeded = NotGranted" "Stream B.5"
+Assert-Todo "AuthorizationDecision shape — Allow + NotGranted only; Denied variant exists unused" "Stream B.6"
+
+Write-Host ""
+Write-Host "Stream C — OPC UA operation wiring"
+Assert-Todo "Every operation wired — Browse/Read/Write/HistoryRead/HistoryUpdate/CreateMonitoredItems/TransferSubscriptions/Call/Ack/Confirm/Shelve" "Stream C.1-C.7"
+Assert-Todo "HistoryRead uses its own flag — Read+no-HistoryRead denies HistoryRead" "Stream C.3"
+Assert-Todo "Mixed-batch semantics — 3 allowed + 2 denied returns per-item status, no coarse failure" "Stream C.6"
+Assert-Todo "Browse ancestor visibility — deep grant implies ancestor browse; denied ancestors filter" "Stream C.7"
+Assert-Todo "Subscription re-authorization — revoked grant surfaces BadUserAccessDenied in one publish" "Stream C.5"
+
+Write-Host ""
+Write-Host "Stream D — Admin UI + SignalR invalidation"
+Assert-Todo "SignalR invalidation — sp_PublishGeneration pushes PermissionTrieCache invalidate < 2 s" "Stream D.4"
+
+Write-Host ""
+Write-Host "Cross-cutting"
+Assert-Todo "No test-count regression — dotnet test ZB.MOM.WW.OtOpcUa.slnx count ≥ pre-Phase-6.2 baseline" "Final exit-gate"
+
+Write-Host ""
+if ($script:failures -eq 0) {
+    Write-Host "Phase 6.2 compliance: scaffold-mode PASS (all checks TODO)" -ForegroundColor Green
+    exit 0
+}
+Write-Host "Phase 6.2 compliance: $script:failures FAIL(s)" -ForegroundColor Red
+exit 1

scripts/compliance/phase-6-3-compliance.ps1

@@ -0,0 +1,85 @@
+<#
+.SYNOPSIS
+    Phase 6.3 exit-gate compliance check — stub. Each `Assert-*` either passes
+    (Write-Host green) or throws. Non-zero exit = fail.
+
+.DESCRIPTION
+    Validates Phase 6.3 (Redundancy runtime) completion. Checks enumerated in
+    `docs/v2/implementation/phase-6-3-redundancy-runtime.md`
+    §"Compliance Checks (run at exit gate)".
+
+    Current status: SCAFFOLD. Every check writes a TODO line and does NOT throw.
+    Each implementation task in Phase 6.3 is responsible for replacing its TODO
+    with a real check before closing that task.
+
+.NOTES
+    Usage:  pwsh ./scripts/compliance/phase-6-3-compliance.ps1
+    Exit:   0 = all checks passed (or are still TODO); non-zero = explicit fail
+#>
+[CmdletBinding()]
+param()
+
+$ErrorActionPreference = 'Stop'
+$script:failures = 0
+
+function Assert-Todo {
+    param([string]$Check, [string]$ImplementationTask)
+    Write-Host "  [TODO] $Check  (implement during $ImplementationTask)" -ForegroundColor Yellow
+}
+
+function Assert-Pass {
+    param([string]$Check)
+    Write-Host "  [PASS] $Check" -ForegroundColor Green
+}
+
+function Assert-Fail {
+    param([string]$Check, [string]$Reason)
+    Write-Host "  [FAIL] $Check — $Reason" -ForegroundColor Red
+    $script:failures++
+}
+
+Write-Host ""
+Write-Host "=== Phase 6.3 compliance — Redundancy runtime ===" -ForegroundColor Cyan
+Write-Host ""
+
+Write-Host "Stream A — Topology loader"
+Assert-Todo "Transparent-mode rejection — sp_PublishGeneration blocks RedundancyMode=Transparent" "Stream A.3"
+
+Write-Host ""
+Write-Host "Stream B — Peer probe + ServiceLevel calculator"
+Assert-Todo "OPC UA band compliance — 0=Maintenance / 1=NoData reserved; operational 2..255" "Stream B.2"
+Assert-Todo "Authoritative-Primary ServiceLevel = 255" "Stream B.2"
+Assert-Todo "Isolated-Primary (peer unreachable, self serving) = 230" "Stream B.2"
+Assert-Todo "Primary-Mid-Apply = 200" "Stream B.2"
+Assert-Todo "Recovering-Primary = 180 with dwell + publish witness enforced" "Stream B.2"
+Assert-Todo "Authoritative-Backup = 100" "Stream B.2"
+Assert-Todo "Isolated-Backup (primary unreachable) = 80 — no auto-promote" "Stream B.2"
+Assert-Todo "InvalidTopology = 2 — >1 Primary self-demotes both nodes" "Stream B.2"
+Assert-Todo "UaHealthProbe authority — HTTP-200 + UA-down peer treated as UA-unhealthy" "Stream B.1"
+
+Write-Host ""
+Write-Host "Stream C — OPC UA node wiring"
+Assert-Todo "ServerUriArray — returns self + peer URIs, self first" "Stream C.2"
+Assert-Todo "Client.CLI cutover — primary halt triggers reconnect to backup via ServerUriArray" "Stream C.4"
+
+Write-Host ""
+Write-Host "Stream D — Apply-lease + publish fencing"
+Assert-Todo "Apply-lease disposal — leases close on exception, cancellation, watchdog timeout" "Stream D.2"
+Assert-Todo "Role transition via operator publish — no restart; both nodes flip ServiceLevel on publish confirm" "Stream D.3"
+
+Write-Host ""
+Write-Host "Stream F — Interop matrix"
+Assert-Todo "Client interoperability matrix — Ignition 8.1/8.3 / Kepware / Aveva OI Gateway findings documented" "Stream F.1-F.2"
+Assert-Todo "Galaxy MXAccess failover — primary kill; Galaxy consumer reconnects within session-timeout budget" "Stream F.3"
+
+Write-Host ""
+Write-Host "Cross-cutting"
+Assert-Todo "No regression in driver test suites; /healthz reachable under redundancy load" "Final exit-gate"
+
+Write-Host ""
+if ($script:failures -eq 0) {
+    Write-Host "Phase 6.3 compliance: scaffold-mode PASS (all checks TODO)" -ForegroundColor Green
+    exit 0
+}
+Write-Host "Phase 6.3 compliance: $script:failures FAIL(s)" -ForegroundColor Red
+exit 1

scripts/compliance/phase-6-4-compliance.ps1

@@ -0,0 +1,83 @@
+<#
+.SYNOPSIS
+    Phase 6.4 exit-gate compliance check — stub. Each `Assert-*` either passes
+    (Write-Host green) or throws. Non-zero exit = fail.
+
+.DESCRIPTION
+    Validates Phase 6.4 (Admin UI completion) completion. Checks enumerated in
+    `docs/v2/implementation/phase-6-4-admin-ui-completion.md`
+    §"Compliance Checks (run at exit gate)".
+
+    Current status: SCAFFOLD. Every check writes a TODO line and does NOT throw.
+    Each implementation task in Phase 6.4 is responsible for replacing its TODO
+    with a real check before closing that task.
+
+.NOTES
+    Usage:  pwsh ./scripts/compliance/phase-6-4-compliance.ps1
+    Exit:   0 = all checks passed (or are still TODO); non-zero = explicit fail
+#>
+[CmdletBinding()]
+param()
+
+$ErrorActionPreference = 'Stop'
+$script:failures = 0
+
+function Assert-Todo {
+    param([string]$Check, [string]$ImplementationTask)
+    Write-Host "  [TODO] $Check  (implement during $ImplementationTask)" -ForegroundColor Yellow
+}
+
+function Assert-Pass {
+    param([string]$Check)
+    Write-Host "  [PASS] $Check" -ForegroundColor Green
+}
+
+function Assert-Fail {
+    param([string]$Check, [string]$Reason)
+    Write-Host "  [FAIL] $Check — $Reason" -ForegroundColor Red
+    $script:failures++
+}
+
+Write-Host ""
+Write-Host "=== Phase 6.4 compliance — Admin UI completion ===" -ForegroundColor Cyan
+Write-Host ""
+
+Write-Host "Stream A — UNS drag/move + impact preview"
+Assert-Todo "UNS drag/move — drag line across areas; modal shows correct impacted-equipment + tag counts" "Stream A.2"
+Assert-Todo "Concurrent-edit safety — session B saves draft mid-preview; session A Confirm returns 409" "Stream A.3 (DraftRevisionToken)"
+Assert-Todo "Cross-cluster drop disabled — actionable toast points to Export/Import" "Stream A.2"
+Assert-Todo "1000-node tree — drag-enter feedback < 100 ms" "Stream A.4"
+
+Write-Host ""
+Write-Host "Stream B — CSV import + staged-import + 5-identifier search"
+Assert-Todo "CSV header version — file missing '# OtOpcUaCsv v1' rejected pre-parse" "Stream B.1"
+Assert-Todo "CSV canonical identifier set — columns match decision #117 exactly" "Stream B.1"
+Assert-Todo "Staged-import atomicity — 10k-row FinaliseImportBatch < 30 s; user-scoped visibility; DropImportBatch rollback" "Stream B.3"
+Assert-Todo "Concurrent import + external reservation — finalize retries with conflict handling; no corruption" "Stream B.3"
+Assert-Todo "5-identifier search ranking — exact > prefix; published > draft for equal scores" "Stream B.4"
+
+Write-Host ""
+Write-Host "Stream C — DiffViewer sections"
+Assert-Todo "Diff viewer section caps — 2000-row subtree-rename summary-only; 'Load full diff' paginates" "Stream C.2"
+
+Write-Host ""
+Write-Host "Stream D — Identification (OPC 40010)"
+Assert-Todo "OPC 40010 field list match — rendered fields match decision #139 exactly; no extras" "Stream D.1"
+Assert-Todo "OPC 40010 exposure — Identification sub-folder shows when non-null; absent when all null" "Stream D.3"
+Assert-Todo "ACL inheritance for Identification — Equipment-grant reads; no-grant denies both" "Stream D.4"
+
+Write-Host ""
+Write-Host "Visual compliance"
+Assert-Todo "Visual parity reviewer — FleetAdmin signoff vs admin-ui.md §Visual-Design; screenshot set checked in under docs/v2/visual-compliance/phase-6-4/" "Visual review"
+
+Write-Host ""
+Write-Host "Cross-cutting"
+Assert-Todo "Full solution dotnet test passes; no test-count regression vs pre-Phase-6.4 baseline" "Final exit-gate"
+
+Write-Host ""
+if ($script:failures -eq 0) {
+    Write-Host "Phase 6.4 compliance: scaffold-mode PASS (all checks TODO)" -ForegroundColor Green
+    exit 0
+}
+Write-Host "Phase 6.4 compliance: $script:failures FAIL(s)" -ForegroundColor Red
+exit 1

src/ZB.MOM.WW.OtOpcUa.Core.Abstractions/DriverAttributeInfo.cs

@@ -25,6 +25,14 @@ namespace ZB.MOM.WW.OtOpcUa.Core.Abstractions;
 ///     OPC UA <c>AlarmConditionState</c> when true. Defaults to false so existing non-Galaxy
 ///     drivers aren't forced to flow a flag they don't produce.
 /// </param>
+/// <param name="WriteIdempotent">
+///     True when a timed-out or failed write to this attribute is safe to replay. Per
+///     <c>docs/v2/plan.md</c> decisions #44, #45, #143 — writes are NOT auto-retried by default
+///     because replaying a pulse / alarm-ack / counter-increment / recipe-step advance can
+///     duplicate field actions. Drivers flag only tags whose semantics make retry safe
+///     (holding registers with level-set values, set-point writes to analog tags) — the
+///     capability invoker respects this flag when deciding whether to apply Polly retry.
+/// </param>
 public sealed record DriverAttributeInfo(
     string FullName,
     DriverDataType DriverDataType,
@@ -32,4 +40,5 @@ public sealed record DriverAttributeInfo(
     uint? ArrayDim,
     SecurityClassification SecurityClass,
     bool IsHistorized,
-    bool IsAlarm = false);
+    bool IsAlarm = false,
+    bool WriteIdempotent = false);

src/ZB.MOM.WW.OtOpcUa.Core.Abstractions/DriverCapability.cs

@@ -0,0 +1,42 @@
+namespace ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+/// <summary>
+///     Enumerates the driver-capability surface points guarded by Phase 6.1 resilience pipelines.
+///     Each value corresponds to one method (or tightly-related method group) on the
+///     <c>Core.Abstractions</c> capability interfaces (<see cref="IReadable"/>, <see cref="IWritable"/>,
+///     <see cref="ITagDiscovery"/>, <see cref="ISubscribable"/>, <see cref="IHostConnectivityProbe"/>,
+///     <see cref="IAlarmSource"/>, <see cref="IHistoryProvider"/>).
+/// </summary>
+/// <remarks>
+///     Per <c>docs/v2/plan.md</c> decision #143 (per-capability retry policy): Read / HistoryRead /
+///     Discover / Probe / AlarmSubscribe auto-retry; <see cref="Write"/> does NOT retry unless the
+///     tag-definition carries <see cref="WriteIdempotentAttribute"/>. Alarm-acknowledge is treated
+///     as a write for retry semantics (an alarm-ack is not idempotent at the plant-floor acknowledgement
+///     level even if the OPC UA spec permits re-issue).
+/// </remarks>
+public enum DriverCapability
+{
+    /// <summary>Batch <see cref="IReadable.ReadAsync"/>. Retries by default.</summary>
+    Read,
+
+    /// <summary>Batch <see cref="IWritable.WriteAsync"/>. Does not retry unless tag is <see cref="WriteIdempotentAttribute">idempotent</see>.</summary>
+    Write,
+
+    /// <summary><see cref="ITagDiscovery.DiscoverAsync"/>. Retries by default.</summary>
+    Discover,
+
+    /// <summary><see cref="ISubscribable.SubscribeAsync"/> and unsubscribe. Retries by default.</summary>
+    Subscribe,
+
+    /// <summary><see cref="IHostConnectivityProbe"/> probe loop. Retries by default.</summary>
+    Probe,
+
+    /// <summary><see cref="IAlarmSource.SubscribeAlarmsAsync"/>. Retries by default.</summary>
+    AlarmSubscribe,
+
+    /// <summary><see cref="IAlarmSource.AcknowledgeAsync"/>. Does NOT retry — ack is a write-shaped operation (decision #143).</summary>
+    AlarmAcknowledge,
+
+    /// <summary><see cref="IHistoryProvider"/> reads (Raw/Processed/AtTime/Events). Retries by default.</summary>
+    HistoryRead,
+}

src/ZB.MOM.WW.OtOpcUa.Core.Abstractions/DriverTier.cs

@@ -0,0 +1,34 @@
+namespace ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+/// <summary>
+///     Stability tier of a driver type. Determines which cross-cutting runtime protections
+///     apply — per-tier retry defaults, memory-tracking thresholds, and whether out-of-process
+///     supervision with process-level recycle is in play.
+/// </summary>
+/// <remarks>
+///     Per <c>docs/v2/driver-stability.md</c> §2-4 and <c>docs/v2/plan.md</c> decisions #63-74.
+///
+///     <list type="bullet">
+///         <item><b>A</b> — managed, known-good SDK; low blast radius. In-process. Fast retries.
+///             Examples: OPC UA Client (OPCFoundation stack), S7 (S7NetPlus).</item>
+///         <item><b>B</b> — native or semi-trusted SDK with an in-process footprint. Examples: Modbus.</item>
+///         <item><b>C</b> — unmanaged SDK with COM/STA constraints, leak risk, or other out-of-process
+///             requirements. Must run as a separate Host process behind a Proxy with a supervisor that
+///             can recycle the process on hard-breach. Example: Galaxy (MXAccess COM).</item>
+///     </list>
+///
+///     <para>Process-kill protections (<c>MemoryRecycle</c>, <c>ScheduledRecycleScheduler</c>) are
+///     Tier C only per decisions #73-74 and #145 — killing an in-process Tier A/B driver also kills
+///     every OPC UA session and every co-hosted driver, blast-radius worse than the leak.</para>
+/// </remarks>
+public enum DriverTier
+{
+    /// <summary>Managed SDK, in-process, low blast radius.</summary>
+    A,
+
+    /// <summary>Native or semi-trusted SDK, in-process.</summary>
+    B,
+
+    /// <summary>Unmanaged SDK, out-of-process required with Proxy+Host+Supervisor.</summary>
+    C,
+}

src/ZB.MOM.WW.OtOpcUa.Core.Abstractions/WriteIdempotentAttribute.cs

@@ -0,0 +1,19 @@
+namespace ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+/// <summary>
+///     Opts a tag-definition record into auto-retry on <see cref="IWritable.WriteAsync"/> failures.
+///     Absence of this attribute means writes are <b>not</b> retried — a timed-out write may have
+///     already succeeded at the device, and replaying pulses, alarm acks, counter increments, or
+///     recipe-step advances can duplicate irreversible field actions.
+/// </summary>
+/// <remarks>
+///     Per <c>docs/v2/plan.md</c> decisions #44, #45, and #143. Applied to tag-definition POCOs
+///     (e.g. <c>ModbusTagDefinition</c>, <c>S7TagDefinition</c>, OPC UA client tag rows) at the
+///     property or record level. The <c>CapabilityInvoker</c> in <c>ZB.MOM.WW.OtOpcUa.Core.Resilience</c>
+///     reads this attribute via reflection once at driver-init time and caches the result; no
+///     per-write reflection cost.
+/// </remarks>
+[AttributeUsage(AttributeTargets.Property | AttributeTargets.Class | AttributeTargets.Struct, AllowMultiple = false, Inherited = true)]
+public sealed class WriteIdempotentAttribute : Attribute
+{
+}

src/ZB.MOM.WW.OtOpcUa.Core/Resilience/CapabilityInvoker.cs

@@ -0,0 +1,106 @@
+using Polly;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+/// <summary>
+///     Executes driver-capability calls through a shared Polly pipeline. One invoker per
+///     <c>(DriverInstance, IDriver)</c> pair; the underlying <see cref="DriverResiliencePipelineBuilder"/>
+///     is process-singleton so all invokers share its cache.
+/// </summary>
+/// <remarks>
+///     Per <c>docs/v2/plan.md</c> decisions #143-144 and Phase 6.1 Stream A.3. The server's dispatch
+///     layer routes every capability call (<c>IReadable.ReadAsync</c>, <c>IWritable.WriteAsync</c>,
+///     <c>ITagDiscovery.DiscoverAsync</c>, <c>ISubscribable.SubscribeAsync/UnsubscribeAsync</c>,
+///     <c>IHostConnectivityProbe</c> probe loop, <c>IAlarmSource.SubscribeAlarmsAsync/AcknowledgeAsync</c>,
+///     and all four <c>IHistoryProvider</c> reads) through this invoker.
+/// </remarks>
+public sealed class CapabilityInvoker
+{
+    private readonly DriverResiliencePipelineBuilder _builder;
+    private readonly string _driverInstanceId;
+    private readonly Func<DriverResilienceOptions> _optionsAccessor;
+
+    /// <summary>
+    ///     Construct an invoker for one driver instance.
+    /// </summary>
+    /// <param name="builder">Shared, process-singleton pipeline builder.</param>
+    /// <param name="driverInstanceId">The <c>DriverInstance.Id</c> column value.</param>
+    /// <param name="optionsAccessor">
+    ///     Snapshot accessor for the current resilience options. Invoked per call so Admin-edit +
+    ///     pipeline-invalidate can take effect without restarting the invoker.
+    /// </param>
+    public CapabilityInvoker(
+        DriverResiliencePipelineBuilder builder,
+        string driverInstanceId,
+        Func<DriverResilienceOptions> optionsAccessor)
+    {
+        ArgumentNullException.ThrowIfNull(builder);
+        ArgumentNullException.ThrowIfNull(optionsAccessor);
+
+        _builder = builder;
+        _driverInstanceId = driverInstanceId;
+        _optionsAccessor = optionsAccessor;
+    }
+
+    /// <summary>Execute a capability call returning a value, honoring the per-capability pipeline.</summary>
+    /// <typeparam name="TResult">Return type of the underlying driver call.</typeparam>
+    public async ValueTask<TResult> ExecuteAsync<TResult>(
+        DriverCapability capability,
+        string hostName,
+        Func<CancellationToken, ValueTask<TResult>> callSite,
+        CancellationToken cancellationToken)
+    {
+        ArgumentNullException.ThrowIfNull(callSite);
+
+        var pipeline = ResolvePipeline(capability, hostName);
+        return await pipeline.ExecuteAsync(callSite, cancellationToken).ConfigureAwait(false);
+    }
+
+    /// <summary>Execute a void-returning capability call, honoring the per-capability pipeline.</summary>
+    public async ValueTask ExecuteAsync(
+        DriverCapability capability,
+        string hostName,
+        Func<CancellationToken, ValueTask> callSite,
+        CancellationToken cancellationToken)
+    {
+        ArgumentNullException.ThrowIfNull(callSite);
+
+        var pipeline = ResolvePipeline(capability, hostName);
+        await pipeline.ExecuteAsync(callSite, cancellationToken).ConfigureAwait(false);
+    }
+
+    /// <summary>
+    ///     Execute a <see cref="DriverCapability.Write"/> call honoring <see cref="WriteIdempotentAttribute"/>
+    ///     semantics — if <paramref name="isIdempotent"/> is <c>false</c>, retries are disabled regardless
+    ///     of the tag-level configuration (the pipeline for a non-idempotent write never retries per
+    ///     decisions #44-45). If <c>true</c>, the call runs through the capability's pipeline which may
+    ///     retry when the tier configuration permits.
+    /// </summary>
+    public async ValueTask<TResult> ExecuteWriteAsync<TResult>(
+        string hostName,
+        bool isIdempotent,
+        Func<CancellationToken, ValueTask<TResult>> callSite,
+        CancellationToken cancellationToken)
+    {
+        ArgumentNullException.ThrowIfNull(callSite);
+
+        if (!isIdempotent)
+        {
+            var noRetryOptions = _optionsAccessor() with
+            {
+                CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+                {
+                    [DriverCapability.Write] = _optionsAccessor().Resolve(DriverCapability.Write) with { RetryCount = 0 },
+                },
+            };
+            var pipeline = _builder.GetOrCreate(_driverInstanceId, $"{hostName}::non-idempotent", DriverCapability.Write, noRetryOptions);
+            return await pipeline.ExecuteAsync(callSite, cancellationToken).ConfigureAwait(false);
+        }
+
+        return await ExecuteAsync(DriverCapability.Write, hostName, callSite, cancellationToken).ConfigureAwait(false);
+    }
+
+    private ResiliencePipeline ResolvePipeline(DriverCapability capability, string hostName) =>
+        _builder.GetOrCreate(_driverInstanceId, hostName, capability, _optionsAccessor());
+}

src/ZB.MOM.WW.OtOpcUa.Core/Resilience/DriverResilienceOptions.cs

@@ -0,0 +1,96 @@
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+/// <summary>
+///     Per-tier × per-capability resilience policy configuration for a driver instance.
+///     Bound from <c>DriverInstance.ResilienceConfig</c> JSON (nullable column; null = tier defaults).
+///     Per <c>docs/v2/plan.md</c> decisions #143 and #144.
+/// </summary>
+public sealed record DriverResilienceOptions
+{
+    /// <summary>Tier the owning driver type is registered as; drives the default map.</summary>
+    public required DriverTier Tier { get; init; }
+
+    /// <summary>
+    ///     Per-capability policy overrides. Capabilities absent from this map fall back to
+    ///     <see cref="GetTierDefaults(DriverTier)"/> for the configured <see cref="Tier"/>.
+    /// </summary>
+    public IReadOnlyDictionary<DriverCapability, CapabilityPolicy> CapabilityPolicies { get; init; }
+        = new Dictionary<DriverCapability, CapabilityPolicy>();
+
+    /// <summary>Bulkhead (max concurrent in-flight calls) for every capability. Default 32.</summary>
+    public int BulkheadMaxConcurrent { get; init; } = 32;
+
+    /// <summary>
+    ///     Bulkhead queue depth. Zero = no queueing; overflow fails fast with
+    ///     <c>BulkheadRejectedException</c>. Default 64.
+    /// </summary>
+    public int BulkheadMaxQueue { get; init; } = 64;
+
+    /// <summary>
+    ///     Look up the effective policy for a capability, falling back to tier defaults when no
+    ///     override is configured. Never returns null.
+    /// </summary>
+    public CapabilityPolicy Resolve(DriverCapability capability)
+    {
+        if (CapabilityPolicies.TryGetValue(capability, out var policy))
+            return policy;
+
+        var defaults = GetTierDefaults(Tier);
+        return defaults[capability];
+    }
+
+    /// <summary>
+    ///     Per-tier per-capability default policy table, per decisions #143-144 and the Phase 6.1
+    ///     Stream A.2 specification. Retries skipped on <see cref="DriverCapability.Write"/> and
+    ///     <see cref="DriverCapability.AlarmAcknowledge"/> regardless of tier.
+    /// </summary>
+    public static IReadOnlyDictionary<DriverCapability, CapabilityPolicy> GetTierDefaults(DriverTier tier) =>
+        tier switch
+        {
+            DriverTier.A => new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Read]             = new(TimeoutSeconds: 2,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.Write]            = new(TimeoutSeconds: 2,  RetryCount: 0, BreakerFailureThreshold: 5),
+                [DriverCapability.Discover]         = new(TimeoutSeconds: 30, RetryCount: 2, BreakerFailureThreshold: 3),
+                [DriverCapability.Subscribe]        = new(TimeoutSeconds: 5,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.Probe]            = new(TimeoutSeconds: 2,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.AlarmSubscribe]   = new(TimeoutSeconds: 5,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.AlarmAcknowledge] = new(TimeoutSeconds: 5,  RetryCount: 0, BreakerFailureThreshold: 5),
+                [DriverCapability.HistoryRead]      = new(TimeoutSeconds: 30, RetryCount: 2, BreakerFailureThreshold: 5),
+            },
+            DriverTier.B => new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Read]             = new(TimeoutSeconds: 4,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.Write]            = new(TimeoutSeconds: 4,  RetryCount: 0, BreakerFailureThreshold: 5),
+                [DriverCapability.Discover]         = new(TimeoutSeconds: 60, RetryCount: 2, BreakerFailureThreshold: 3),
+                [DriverCapability.Subscribe]        = new(TimeoutSeconds: 8,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.Probe]            = new(TimeoutSeconds: 4,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.AlarmSubscribe]   = new(TimeoutSeconds: 8,  RetryCount: 3, BreakerFailureThreshold: 5),
+                [DriverCapability.AlarmAcknowledge] = new(TimeoutSeconds: 8,  RetryCount: 0, BreakerFailureThreshold: 5),
+                [DriverCapability.HistoryRead]      = new(TimeoutSeconds: 60, RetryCount: 2, BreakerFailureThreshold: 5),
+            },
+            DriverTier.C => new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Read]             = new(TimeoutSeconds: 10,  RetryCount: 1, BreakerFailureThreshold: 0),
+                [DriverCapability.Write]            = new(TimeoutSeconds: 10,  RetryCount: 0, BreakerFailureThreshold: 0),
+                [DriverCapability.Discover]         = new(TimeoutSeconds: 120, RetryCount: 1, BreakerFailureThreshold: 0),
+                [DriverCapability.Subscribe]        = new(TimeoutSeconds: 15,  RetryCount: 1, BreakerFailureThreshold: 0),
+                [DriverCapability.Probe]            = new(TimeoutSeconds: 10,  RetryCount: 1, BreakerFailureThreshold: 0),
+                [DriverCapability.AlarmSubscribe]   = new(TimeoutSeconds: 15,  RetryCount: 1, BreakerFailureThreshold: 0),
+                [DriverCapability.AlarmAcknowledge] = new(TimeoutSeconds: 15,  RetryCount: 0, BreakerFailureThreshold: 0),
+                [DriverCapability.HistoryRead]      = new(TimeoutSeconds: 120, RetryCount: 1, BreakerFailureThreshold: 0),
+            },
+            _ => throw new ArgumentOutOfRangeException(nameof(tier), tier, $"No default policy table defined for tier {tier}."),
+        };
+}
+
+/// <summary>Policy for one capability on one driver instance.</summary>
+/// <param name="TimeoutSeconds">Per-call timeout (wraps the inner Polly execution).</param>
+/// <param name="RetryCount">Number of retry attempts after the first failure; zero = no retry.</param>
+/// <param name="BreakerFailureThreshold">
+///     Consecutive-failure count that opens the circuit breaker; zero = no breaker
+///     (Tier C uses the supervisor's process-level breaker instead, per decision #68).
+/// </param>
+public sealed record CapabilityPolicy(int TimeoutSeconds, int RetryCount, int BreakerFailureThreshold);

src/ZB.MOM.WW.OtOpcUa.Core/Resilience/DriverResiliencePipelineBuilder.cs

@@ -0,0 +1,118 @@
+using System.Collections.Concurrent;
+using Polly;
+using Polly.CircuitBreaker;
+using Polly.Retry;
+using Polly.Timeout;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+/// <summary>
+///     Builds and caches Polly resilience pipelines keyed on
+///     <c>(DriverInstanceId, HostName, DriverCapability)</c>. One dead PLC behind a multi-device
+///     driver cannot open the circuit breaker for healthy sibling hosts.
+/// </summary>
+/// <remarks>
+///     Per <c>docs/v2/plan.md</c> decision #144 (per-device isolation). Composition from outside-in:
+///     <b>Timeout → Retry (when capability permits) → Circuit Breaker (when tier permits) → Bulkhead</b>.
+///
+///     <para>Pipeline resolution is lock-free on the hot path: the inner
+///     <see cref="ConcurrentDictionary{TKey,TValue}"/> caches a <see cref="ResiliencePipeline"/> per key;
+///     first-call cost is one <see cref="ResiliencePipelineBuilder"/>.Build. Thereafter reads are O(1).</para>
+/// </remarks>
+public sealed class DriverResiliencePipelineBuilder
+{
+    private readonly ConcurrentDictionary<PipelineKey, ResiliencePipeline> _pipelines = new();
+    private readonly TimeProvider _timeProvider;
+
+    /// <summary>Construct with the ambient clock (use <see cref="TimeProvider.System"/> in prod).</summary>
+    public DriverResiliencePipelineBuilder(TimeProvider? timeProvider = null)
+    {
+        _timeProvider = timeProvider ?? TimeProvider.System;
+    }
+
+    /// <summary>
+    ///     Get or build the pipeline for a given <c>(driver instance, host, capability)</c> triple.
+    ///     Calls with the same key + same options reuse the same pipeline instance; the first caller
+    ///     wins if a race occurs (both pipelines would be behaviourally identical).
+    /// </summary>
+    /// <param name="driverInstanceId">DriverInstance primary key — opaque to this layer.</param>
+    /// <param name="hostName">
+    ///     Host the call targets. For single-host drivers (Galaxy, some OPC UA Client configs) pass the
+    ///     driver's canonical host string. For multi-host drivers (Modbus with N PLCs), pass the
+    ///     specific PLC so one dead PLC doesn't poison healthy siblings.
+    /// </param>
+    /// <param name="capability">Which capability surface is being called.</param>
+    /// <param name="options">Per-driver-instance options (tier + per-capability overrides).</param>
+    public ResiliencePipeline GetOrCreate(
+        string driverInstanceId,
+        string hostName,
+        DriverCapability capability,
+        DriverResilienceOptions options)
+    {
+        ArgumentNullException.ThrowIfNull(options);
+        ArgumentException.ThrowIfNullOrWhiteSpace(hostName);
+
+        var key = new PipelineKey(driverInstanceId, hostName, capability);
+        return _pipelines.GetOrAdd(key, static (_, state) => Build(state.capability, state.options, state.timeProvider),
+            (capability, options, timeProvider: _timeProvider));
+    }
+
+    /// <summary>Drop cached pipelines for one driver instance (e.g. on ResilienceConfig change). Test + Admin-reload use.</summary>
+    public int Invalidate(string driverInstanceId)
+    {
+        var removed = 0;
+        foreach (var key in _pipelines.Keys)
+        {
+            if (key.DriverInstanceId == driverInstanceId && _pipelines.TryRemove(key, out _))
+                removed++;
+        }
+        return removed;
+    }
+
+    /// <summary>Snapshot of the current number of cached pipelines. For diagnostics only.</summary>
+    public int CachedPipelineCount => _pipelines.Count;
+
+    private static ResiliencePipeline Build(
+        DriverCapability capability,
+        DriverResilienceOptions options,
+        TimeProvider timeProvider)
+    {
+        var policy = options.Resolve(capability);
+        var builder = new ResiliencePipelineBuilder { TimeProvider = timeProvider };
+
+        builder.AddTimeout(new TimeoutStrategyOptions
+        {
+            Timeout = TimeSpan.FromSeconds(policy.TimeoutSeconds),
+        });
+
+        if (policy.RetryCount > 0)
+        {
+            builder.AddRetry(new RetryStrategyOptions
+            {
+                MaxRetryAttempts = policy.RetryCount,
+                BackoffType = DelayBackoffType.Exponential,
+                UseJitter = true,
+                Delay = TimeSpan.FromMilliseconds(100),
+                MaxDelay = TimeSpan.FromSeconds(5),
+                ShouldHandle = new PredicateBuilder().Handle<Exception>(ex => ex is not OperationCanceledException),
+            });
+        }
+
+        if (policy.BreakerFailureThreshold > 0)
+        {
+            builder.AddCircuitBreaker(new CircuitBreakerStrategyOptions
+            {
+                FailureRatio = 1.0,
+                MinimumThroughput = policy.BreakerFailureThreshold,
+                SamplingDuration = TimeSpan.FromSeconds(30),
+                BreakDuration = TimeSpan.FromSeconds(15),
+                ShouldHandle = new PredicateBuilder().Handle<Exception>(ex => ex is not OperationCanceledException),
+            });
+        }
+
+        return builder.Build();
+    }
+
+    private readonly record struct PipelineKey(string DriverInstanceId, string HostName, DriverCapability Capability);
+}

src/ZB.MOM.WW.OtOpcUa.Core/ZB.MOM.WW.OtOpcUa.Core.csproj

@@ -16,6 +16,10 @@
         <ProjectReference Include="..\ZB.MOM.WW.OtOpcUa.Configuration\ZB.MOM.WW.OtOpcUa.Configuration.csproj"/>
     </ItemGroup>
 
+    <ItemGroup>
+        <PackageReference Include="Polly.Core" Version="8.6.6"/>
+    </ItemGroup>
+
     <ItemGroup>
         <InternalsVisibleTo Include="ZB.MOM.WW.OtOpcUa.Core.Tests"/>
     </ItemGroup>

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

@@ -115,7 +115,8 @@ public sealed class ModbusDriver(ModbusDriverOptions options, string driverInsta
                 ArrayDim: null,
                 SecurityClass: t.Writable ? SecurityClassification.Operate : SecurityClassification.ViewOnly,
                 IsHistorized: false,
-                IsAlarm: false));
+                IsAlarm: false,
+                WriteIdempotent: t.WriteIdempotent));
         }
         return Task.CompletedTask;
     }

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

@@ -92,6 +92,14 @@ public sealed class ModbusProbeOptions
 ///     AutomationDirect DirectLOGIC (DL205/DL260) and a few legacy families pack the first
 ///     character in the low byte instead — see <c>docs/v2/dl205.md</c> §strings.
 /// </param>
+/// <param name="WriteIdempotent">
+///     Per <c>docs/v2/plan.md</c> decisions #44, #45, #143 — flag a tag as safe to replay on
+///     write timeout / failure. Default <c>false</c>; writes do not auto-retry. Safe candidates:
+///     holding-register set-points for analog values and configuration registers where the same
+///     value can be written again without side-effects. Unsafe: coils that drive edge-triggered
+///     actions (pulse outputs), counter-increment addresses on PLCs that treat writes as deltas,
+///     any BCD / counter register where repeat-writes advance state.
+/// </param>
 public sealed record ModbusTagDefinition(
     string Name,
     ModbusRegion Region,
@@ -101,7 +109,8 @@ public sealed record ModbusTagDefinition(
     ModbusByteOrder ByteOrder = ModbusByteOrder.BigEndian,
     byte BitIndex = 0,
     ushort StringLength = 0,
-    ModbusStringByteOrder StringByteOrder = ModbusStringByteOrder.HighByteFirst);
+    ModbusStringByteOrder StringByteOrder = ModbusStringByteOrder.HighByteFirst,
+    bool WriteIdempotent = false);
 
 public enum ModbusRegion { Coils, DiscreteInputs, InputRegisters, HoldingRegisters }
 

src/ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient/OpcUaClientDriver.cs

@@ -27,8 +27,15 @@ namespace ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient;
 ///     </para>
 /// </remarks>
 public sealed class OpcUaClientDriver(OpcUaClientDriverOptions options, string driverInstanceId)
-    : IDriver, ITagDiscovery, IReadable, IWritable, ISubscribable, IHostConnectivityProbe, IDisposable, IAsyncDisposable
+    : IDriver, ITagDiscovery, IReadable, IWritable, ISubscribable, IHostConnectivityProbe, IAlarmSource, IHistoryProvider, IDisposable, IAsyncDisposable
 {
+    // ---- IAlarmSource state ----
+
+    private readonly System.Collections.Concurrent.ConcurrentDictionary<long, RemoteAlarmSubscription> _alarmSubscriptions = new();
+    private long _nextAlarmSubscriptionId;
+
+    public event EventHandler<AlarmEventArgs>? OnAlarmEvent;
+
     // ---- ISubscribable + IHostConnectivityProbe state ----
 
     private readonly System.Collections.Concurrent.ConcurrentDictionary<long, RemoteSubscription> _subscriptions = new();
@@ -395,6 +402,13 @@ public sealed class OpcUaClientDriver(OpcUaClientDriverOptions options, string d
         }
         _subscriptions.Clear();
 
+        foreach (var ras in _alarmSubscriptions.Values)
+        {
+            try { await ras.Subscription.DeleteAsync(silent: true, cancellationToken).ConfigureAwait(false); }
+            catch { /* best-effort */ }
+        }
+        _alarmSubscriptions.Clear();
+
         // Abort any in-flight reconnect attempts before touching the session — BeginReconnect's
         // retry loop holds a reference to the current session and would fight Session.CloseAsync
         // if left spinning.
@@ -936,6 +950,325 @@ public sealed class OpcUaClientDriver(OpcUaClientDriverOptions options, string d
         public string DiagnosticId => $"opcua-sub-{Id}";
     }
 
+    // ---- IAlarmSource ----
+
+    /// <summary>
+    ///     Field positions in the EventFilter SelectClauses below. Used to index into the
+    ///     <c>EventFieldList.EventFields</c> Variant collection when an event arrives.
+    /// </summary>
+    private const int AlarmFieldEventId = 0;
+    private const int AlarmFieldEventType = 1;
+    private const int AlarmFieldSourceNode = 2;
+    private const int AlarmFieldMessage = 3;
+    private const int AlarmFieldSeverity = 4;
+    private const int AlarmFieldTime = 5;
+    private const int AlarmFieldConditionId = 6;
+
+    public async Task<IAlarmSubscriptionHandle> SubscribeAlarmsAsync(
+        IReadOnlyList<string> sourceNodeIds, CancellationToken cancellationToken)
+    {
+        var session = RequireSession();
+        var id = Interlocked.Increment(ref _nextAlarmSubscriptionId);
+        var handle = new OpcUaAlarmSubscriptionHandle(id);
+
+        // Pre-resolve the source-node filter set so the per-event notification handler can
+        // match in O(1) without re-parsing on every event.
+        var sourceFilter = new HashSet<string>(sourceNodeIds, StringComparer.Ordinal);
+
+        var subscription = new Subscription(telemetry: null!, new SubscriptionOptions
+        {
+            DisplayName = $"opcua-alarm-sub-{id}",
+            PublishingInterval = 500,   // 500ms — alarms don't need fast polling; the server pushes
+            KeepAliveCount = 10,
+            LifetimeCount = 1000,
+            MaxNotificationsPerPublish = 0,
+            PublishingEnabled = true,
+            Priority = 0,
+            TimestampsToReturn = TimestampsToReturn.Both,
+        });
+
+        // EventFilter SelectClauses — pick the standard BaseEventType fields we need to
+        // materialize an AlarmEventArgs. Field positions are indexed by the AlarmField*
+        // constants so the notification handler indexes in O(1) without re-examining the
+        // QualifiedName BrowsePaths.
+        var filter = new EventFilter();
+        void AddField(string browseName) => filter.SelectClauses.Add(new SimpleAttributeOperand
+        {
+            TypeDefinitionId = ObjectTypeIds.BaseEventType,
+            BrowsePath = [new QualifiedName(browseName)],
+            AttributeId = Attributes.Value,
+        });
+        AddField("EventId");
+        AddField("EventType");
+        AddField("SourceNode");
+        AddField("Message");
+        AddField("Severity");
+        AddField("Time");
+        // ConditionId on ConditionType nodes is the branch identifier for
+        // acknowledgeable conditions. Not a BaseEventType field — reach it via the typed path.
+        filter.SelectClauses.Add(new SimpleAttributeOperand
+        {
+            TypeDefinitionId = ObjectTypeIds.ConditionType,
+            BrowsePath = [], // empty path = the condition node itself
+            AttributeId = Attributes.NodeId,
+        });
+
+        await _gate.WaitAsync(cancellationToken).ConfigureAwait(false);
+        try
+        {
+            session.AddSubscription(subscription);
+            await subscription.CreateAsync(cancellationToken).ConfigureAwait(false);
+
+            var eventItem = new MonitoredItem(telemetry: null!, new MonitoredItemOptions
+            {
+                DisplayName = "Server/Events",
+                StartNodeId = ObjectIds.Server,
+                AttributeId = Attributes.EventNotifier,
+                MonitoringMode = MonitoringMode.Reporting,
+                QueueSize = 1000, // deep queue — a server can fire many alarms in bursts
+                DiscardOldest = false,
+                Filter = filter,
+            })
+            {
+                Handle = handle,
+            };
+            eventItem.Notification += (mi, args) => OnEventNotification(handle, sourceFilter, mi, args);
+            subscription.AddItem(eventItem);
+            await subscription.CreateItemsAsync(cancellationToken).ConfigureAwait(false);
+
+            _alarmSubscriptions[id] = new RemoteAlarmSubscription(subscription, handle);
+        }
+        finally { _gate.Release(); }
+
+        return handle;
+    }
+
+    public async Task UnsubscribeAlarmsAsync(IAlarmSubscriptionHandle handle, CancellationToken cancellationToken)
+    {
+        if (handle is not OpcUaAlarmSubscriptionHandle h) return;
+        if (!_alarmSubscriptions.TryRemove(h.Id, out var rs)) return;
+
+        await _gate.WaitAsync(cancellationToken).ConfigureAwait(false);
+        try
+        {
+            try { await rs.Subscription.DeleteAsync(silent: true, cancellationToken).ConfigureAwait(false); }
+            catch { /* best-effort — session may already be gone across a reconnect */ }
+        }
+        finally { _gate.Release(); }
+    }
+
+    public async Task AcknowledgeAsync(
+        IReadOnlyList<AlarmAcknowledgeRequest> acknowledgements, CancellationToken cancellationToken)
+    {
+        // Short-circuit empty batch BEFORE touching the session so callers can pass an empty
+        // list without guarding the size themselves — e.g. a bulk-ack UI that built an empty
+        // list because the filter matched nothing.
+        if (acknowledgements.Count == 0) return;
+        var session = RequireSession();
+
+        // OPC UA A&C: call the AcknowledgeableConditionType.Acknowledge method on each
+        // condition node with EventId + Comment arguments. CallAsync accepts a batch —
+        // one CallMethodRequest per ack.
+        var callRequests = new CallMethodRequestCollection();
+        foreach (var ack in acknowledgements)
+        {
+            if (!TryParseNodeId(session, ack.ConditionId, out var conditionId)) continue;
+            callRequests.Add(new CallMethodRequest
+            {
+                ObjectId = conditionId,
+                MethodId = MethodIds.AcknowledgeableConditionType_Acknowledge,
+                InputArguments = [
+                    new Variant(Array.Empty<byte>()),  // EventId — server-side best-effort; empty resolves to 'most recent'
+                    new Variant(new LocalizedText(ack.Comment ?? string.Empty)),
+                ],
+            });
+        }
+
+        if (callRequests.Count == 0) return;
+
+        await _gate.WaitAsync(cancellationToken).ConfigureAwait(false);
+        try
+        {
+            try
+            {
+                _ = await session.CallAsync(
+                    requestHeader: null,
+                    methodsToCall: callRequests,
+                    ct: cancellationToken).ConfigureAwait(false);
+            }
+            catch { /* best-effort — caller's re-ack mechanism catches pathological paths */ }
+        }
+        finally { _gate.Release(); }
+    }
+
+    private void OnEventNotification(
+        OpcUaAlarmSubscriptionHandle handle,
+        HashSet<string> sourceFilter,
+        MonitoredItem item,
+        MonitoredItemNotificationEventArgs args)
+    {
+        if (args.NotificationValue is not EventFieldList efl) return;
+        if (efl.EventFields.Count <= AlarmFieldConditionId) return;
+
+        var sourceNode = efl.EventFields[AlarmFieldSourceNode].Value?.ToString() ?? string.Empty;
+        if (sourceFilter.Count > 0 && !sourceFilter.Contains(sourceNode)) return;
+
+        var eventType = efl.EventFields[AlarmFieldEventType].Value?.ToString() ?? "BaseEventType";
+        var message = (efl.EventFields[AlarmFieldMessage].Value as LocalizedText)?.Text ?? string.Empty;
+        var severity = efl.EventFields[AlarmFieldSeverity].Value is ushort sev ? sev : (ushort)0;
+        var time = efl.EventFields[AlarmFieldTime].Value is DateTime t ? t : DateTime.UtcNow;
+        var conditionId = efl.EventFields[AlarmFieldConditionId].Value?.ToString() ?? string.Empty;
+
+        OnAlarmEvent?.Invoke(this, new AlarmEventArgs(
+            SubscriptionHandle: handle,
+            SourceNodeId: sourceNode,
+            ConditionId: conditionId,
+            AlarmType: eventType,
+            Message: message,
+            Severity: MapSeverity(severity),
+            SourceTimestampUtc: time));
+    }
+
+    /// <summary>
+    ///     Map an OPC UA <c>BaseEventType.Severity</c> (1..1000) to our coarse-grained
+    ///     <see cref="AlarmSeverity"/> bucket. Thresholds match the OPC UA A&amp;C Part 9
+    ///     guidance: 1-200 Low, 201-500 Medium, 501-800 High, 801-1000 Critical.
+    /// </summary>
+    internal static AlarmSeverity MapSeverity(ushort opcSeverity) => opcSeverity switch
+    {
+        <= 200 => AlarmSeverity.Low,
+        <= 500 => AlarmSeverity.Medium,
+        <= 800 => AlarmSeverity.High,
+        _ => AlarmSeverity.Critical,
+    };
+
+    private sealed record RemoteAlarmSubscription(Subscription Subscription, OpcUaAlarmSubscriptionHandle Handle);
+
+    private sealed record OpcUaAlarmSubscriptionHandle(long Id) : IAlarmSubscriptionHandle
+    {
+        public string DiagnosticId => $"opcua-alarm-sub-{Id}";
+    }
+
+    // ---- IHistoryProvider (passthrough to upstream server) ----
+
+    public async Task<Core.Abstractions.HistoryReadResult> ReadRawAsync(
+        string fullReference, DateTime startUtc, DateTime endUtc, uint maxValuesPerNode,
+        CancellationToken cancellationToken)
+    {
+        var details = new ReadRawModifiedDetails
+        {
+            IsReadModified = false,
+            StartTime = startUtc,
+            EndTime = endUtc,
+            NumValuesPerNode = maxValuesPerNode,
+            ReturnBounds = false,
+        };
+        return await ExecuteHistoryReadAsync(fullReference, new ExtensionObject(details), cancellationToken)
+            .ConfigureAwait(false);
+    }
+
+    public async Task<Core.Abstractions.HistoryReadResult> ReadProcessedAsync(
+        string fullReference, DateTime startUtc, DateTime endUtc, TimeSpan interval,
+        HistoryAggregateType aggregate, CancellationToken cancellationToken)
+    {
+        var aggregateId = MapAggregateToNodeId(aggregate);
+        var details = new ReadProcessedDetails
+        {
+            StartTime = startUtc,
+            EndTime = endUtc,
+            ProcessingInterval = interval.TotalMilliseconds,
+            AggregateType = [aggregateId],
+        };
+        return await ExecuteHistoryReadAsync(fullReference, new ExtensionObject(details), cancellationToken)
+            .ConfigureAwait(false);
+    }
+
+    public async Task<Core.Abstractions.HistoryReadResult> ReadAtTimeAsync(
+        string fullReference, IReadOnlyList<DateTime> timestampsUtc, CancellationToken cancellationToken)
+    {
+        var reqTimes = new DateTimeCollection(timestampsUtc);
+        var details = new ReadAtTimeDetails
+        {
+            ReqTimes = reqTimes,
+            UseSimpleBounds = true,
+        };
+        return await ExecuteHistoryReadAsync(fullReference, new ExtensionObject(details), cancellationToken)
+            .ConfigureAwait(false);
+    }
+
+    /// <summary>
+    ///     Shared HistoryRead wire path — used by Raw/Processed/AtTime. Handles NodeId parse,
+    ///     Session.HistoryReadAsync call, Bad-StatusCode passthrough (no translation per §8
+    ///     cascading-quality rule), and HistoryData unwrap into <see cref="DataValueSnapshot"/>.
+    /// </summary>
+    private async Task<Core.Abstractions.HistoryReadResult> ExecuteHistoryReadAsync(
+        string fullReference, ExtensionObject historyReadDetails, CancellationToken ct)
+    {
+        var session = RequireSession();
+        if (!TryParseNodeId(session, fullReference, out var nodeId))
+        {
+            return new Core.Abstractions.HistoryReadResult([], null);
+        }
+
+        var nodesToRead = new HistoryReadValueIdCollection
+        {
+            new HistoryReadValueId { NodeId = nodeId },
+        };
+
+        await _gate.WaitAsync(ct).ConfigureAwait(false);
+        try
+        {
+            var resp = await session.HistoryReadAsync(
+                requestHeader: null,
+                historyReadDetails: historyReadDetails,
+                timestampsToReturn: TimestampsToReturn.Both,
+                releaseContinuationPoints: false,
+                nodesToRead: nodesToRead,
+                ct: ct).ConfigureAwait(false);
+
+            if (resp.Results.Count == 0) return new Core.Abstractions.HistoryReadResult([], null);
+            var r = resp.Results[0];
+
+            // Unwrap HistoryData from the ExtensionObject-encoded payload the SDK returns.
+            // Samples stay in chronological order per OPC UA Part 11; cascading-quality
+            // rule: preserve each DataValue's upstream StatusCode + timestamps verbatim.
+            var samples = new List<DataValueSnapshot>();
+            if (r.HistoryData?.Body is HistoryData hd)
+            {
+                var now = DateTime.UtcNow;
+                foreach (var dv in hd.DataValues)
+                {
+                    samples.Add(new DataValueSnapshot(
+                        Value: dv.Value,
+                        StatusCode: dv.StatusCode.Code,
+                        SourceTimestampUtc: dv.SourceTimestamp == DateTime.MinValue ? null : dv.SourceTimestamp,
+                        ServerTimestampUtc: dv.ServerTimestamp == DateTime.MinValue ? now : dv.ServerTimestamp));
+                }
+            }
+
+            var contPt = r.ContinuationPoint is { Length: > 0 } ? r.ContinuationPoint : null;
+            return new Core.Abstractions.HistoryReadResult(samples, contPt);
+        }
+        finally { _gate.Release(); }
+    }
+
+    /// <summary>Map <see cref="HistoryAggregateType"/> to the OPC UA Part 13 standard aggregate NodeId.</summary>
+    internal static NodeId MapAggregateToNodeId(HistoryAggregateType aggregate) => aggregate switch
+    {
+        HistoryAggregateType.Average => ObjectIds.AggregateFunction_Average,
+        HistoryAggregateType.Minimum => ObjectIds.AggregateFunction_Minimum,
+        HistoryAggregateType.Maximum => ObjectIds.AggregateFunction_Maximum,
+        HistoryAggregateType.Total   => ObjectIds.AggregateFunction_Total,
+        HistoryAggregateType.Count   => ObjectIds.AggregateFunction_Count,
+        _ => throw new ArgumentOutOfRangeException(nameof(aggregate), aggregate, null),
+    };
+
+    // ReadEventsAsync stays at the interface default (throws NotSupportedException) per
+    // IHistoryProvider contract -- the OPC UA Client driver CAN forward HistoryReadEvents,
+    // but the call-site needs an EventFilter SelectClauses surface which the interface
+    // doesn't carry. Landing the event-history passthrough requires extending
+    // IHistoryProvider.ReadEventsAsync with a filter-spec parameter; out of scope for this PR.
+
     // ---- IHostConnectivityProbe ----
 
     /// <summary>

src/ZB.MOM.WW.OtOpcUa.Driver.S7/S7Driver.cs

@@ -341,7 +341,8 @@ public sealed class S7Driver(S7DriverOptions options, string driverInstanceId)
                 ArrayDim: null,
                 SecurityClass: t.Writable ? SecurityClassification.Operate : SecurityClassification.ViewOnly,
                 IsHistorized: false,
-                IsAlarm: false));
+                IsAlarm: false,
+                WriteIdempotent: t.WriteIdempotent));
         }
         return Task.CompletedTask;
     }

src/ZB.MOM.WW.OtOpcUa.Driver.S7/S7DriverOptions.cs

@@ -88,12 +88,20 @@ public sealed class S7ProbeOptions
 /// <param name="DataType">Logical data type — drives the underlying S7.Net read/write width.</param>
 /// <param name="Writable">When true the driver accepts writes for this tag.</param>
 /// <param name="StringLength">For <c>DataType = String</c>: S7-string max length. Default 254 (S7 max).</param>
+/// <param name="WriteIdempotent">
+///     Per <c>docs/v2/plan.md</c> decisions #44, #45, #143 — flag a tag as safe to replay on
+///     write timeout / failure. Default <c>false</c>; writes do not auto-retry. Safe candidates
+///     on S7: DB word/dword set-points holding analog values, configuration DBs where the same
+///     value can be written again without side-effects. Unsafe: M (merker) bits or Q (output)
+///     coils that drive edge-triggered routines in the PLC program.
+/// </param>
 public sealed record S7TagDefinition(
     string Name,
     string Address,
     S7DataType DataType,
     bool Writable = true,
-    int StringLength = 254);
+    int StringLength = 254,
+    bool WriteIdempotent = false);
 
 public enum S7DataType
 {

src/ZB.MOM.WW.OtOpcUa.Server/OpcUa/DriverNodeManager.cs

@@ -3,6 +3,7 @@ using Microsoft.Extensions.Logging;
 using Opc.Ua;
 using Opc.Ua.Server;
 using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
 using ZB.MOM.WW.OtOpcUa.Server.Security;
 using DriverWriteRequest = ZB.MOM.WW.OtOpcUa.Core.Abstractions.WriteRequest;
 // Core.Abstractions defines a type-named HistoryReadResult (driver-side samples + continuation
@@ -33,8 +34,14 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
     private readonly IDriver _driver;
     private readonly IReadable? _readable;
     private readonly IWritable? _writable;
+    private readonly CapabilityInvoker _invoker;
     private readonly ILogger<DriverNodeManager> _logger;
 
+    // Per-variable idempotency flag populated during Variable() registration from
+    // DriverAttributeInfo.WriteIdempotent. Drives ExecuteWriteAsync's retry gating in
+    // OnWriteValue; absent entries default to false (decisions #44, #45, #143).
+    private readonly Dictionary<string, bool> _writeIdempotentByFullRef = new(StringComparer.OrdinalIgnoreCase);
+
     /// <summary>The driver whose address space this node manager exposes.</summary>
     public IDriver Driver => _driver;
 
@@ -53,12 +60,13 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
     private FolderState _currentFolder = null!;
 
     public DriverNodeManager(IServerInternal server, ApplicationConfiguration configuration,
-        IDriver driver, ILogger<DriverNodeManager> logger)
+        IDriver driver, CapabilityInvoker invoker, ILogger<DriverNodeManager> logger)
         : base(server, configuration, namespaceUris: $"urn:OtOpcUa:{driver.DriverInstanceId}")
     {
         _driver = driver;
         _readable = driver as IReadable;
         _writable = driver as IWritable;
+        _invoker = invoker;
         _logger = logger;
     }
 
@@ -148,6 +156,7 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
             AddPredefinedNode(SystemContext, v);
             _variablesByFullRef[attributeInfo.FullName] = v;
             _securityByFullRef[attributeInfo.FullName] = attributeInfo.SecurityClass;
+            _writeIdempotentByFullRef[attributeInfo.FullName] = attributeInfo.WriteIdempotent;
 
             v.OnReadValue = OnReadValue;
             v.OnWriteValue = OnWriteValue;
@@ -188,7 +197,11 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
         try
         {
             var fullRef = node.NodeId.Identifier as string ?? "";
-            var result = _readable.ReadAsync([fullRef], CancellationToken.None).GetAwaiter().GetResult();
+            var result = _invoker.ExecuteAsync(
+                DriverCapability.Read,
+                _driver.DriverInstanceId,
+                async ct => (IReadOnlyList<DataValueSnapshot>)await _readable.ReadAsync([fullRef], ct).ConfigureAwait(false),
+                CancellationToken.None).AsTask().GetAwaiter().GetResult();
             if (result.Count == 0)
             {
                 statusCode = StatusCodes.BadNoData;
@@ -381,9 +394,15 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
 
         try
         {
-            var results = _writable.WriteAsync(
-                [new DriverWriteRequest(fullRef!, value)],
-                CancellationToken.None).GetAwaiter().GetResult();
+            var isIdempotent = _writeIdempotentByFullRef.GetValueOrDefault(fullRef!, false);
+            var capturedValue = value;
+            var results = _invoker.ExecuteWriteAsync(
+                _driver.DriverInstanceId,
+                isIdempotent,
+                async ct => (IReadOnlyList<WriteResult>)await _writable.WriteAsync(
+                    [new DriverWriteRequest(fullRef!, capturedValue)],
+                    ct).ConfigureAwait(false),
+                CancellationToken.None).AsTask().GetAwaiter().GetResult();
             if (results.Count > 0 && results[0].StatusCode != 0)
             {
                 statusCode = results[0].StatusCode;
@@ -465,12 +484,16 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
 
             try
             {
-                var driverResult = History.ReadRawAsync(
-                    fullRef,
-                    details.StartTime,
-                    details.EndTime,
-                    details.NumValuesPerNode,
-                    CancellationToken.None).GetAwaiter().GetResult();
+                var driverResult = _invoker.ExecuteAsync(
+                    DriverCapability.HistoryRead,
+                    _driver.DriverInstanceId,
+                    async ct => await History.ReadRawAsync(
+                        fullRef,
+                        details.StartTime,
+                        details.EndTime,
+                        details.NumValuesPerNode,
+                        ct).ConfigureAwait(false),
+                    CancellationToken.None).AsTask().GetAwaiter().GetResult();
 
                 WriteResult(results, errors, i, StatusCodes.Good,
                     BuildHistoryData(driverResult.Samples), driverResult.ContinuationPoint);
@@ -525,13 +548,17 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
 
             try
             {
-                var driverResult = History.ReadProcessedAsync(
-                    fullRef,
-                    details.StartTime,
-                    details.EndTime,
-                    interval,
-                    aggregate.Value,
-                    CancellationToken.None).GetAwaiter().GetResult();
+                var driverResult = _invoker.ExecuteAsync(
+                    DriverCapability.HistoryRead,
+                    _driver.DriverInstanceId,
+                    async ct => await History.ReadProcessedAsync(
+                        fullRef,
+                        details.StartTime,
+                        details.EndTime,
+                        interval,
+                        aggregate.Value,
+                        ct).ConfigureAwait(false),
+                    CancellationToken.None).AsTask().GetAwaiter().GetResult();
 
                 WriteResult(results, errors, i, StatusCodes.Good,
                     BuildHistoryData(driverResult.Samples), driverResult.ContinuationPoint);
@@ -578,8 +605,11 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
 
             try
             {
-                var driverResult = History.ReadAtTimeAsync(
-                    fullRef, requestedTimes, CancellationToken.None).GetAwaiter().GetResult();
+                var driverResult = _invoker.ExecuteAsync(
+                    DriverCapability.HistoryRead,
+                    _driver.DriverInstanceId,
+                    async ct => await History.ReadAtTimeAsync(fullRef, requestedTimes, ct).ConfigureAwait(false),
+                    CancellationToken.None).AsTask().GetAwaiter().GetResult();
 
                 WriteResult(results, errors, i, StatusCodes.Good,
                     BuildHistoryData(driverResult.Samples), driverResult.ContinuationPoint);
@@ -632,12 +662,16 @@ public sealed class DriverNodeManager : CustomNodeManager2, IAddressSpaceBuilder
 
             try
             {
-                var driverResult = History.ReadEventsAsync(
-                    sourceName: fullRef,
-                    startUtc: details.StartTime,
-                    endUtc: details.EndTime,
-                    maxEvents: maxEvents,
-                    cancellationToken: CancellationToken.None).GetAwaiter().GetResult();
+                var driverResult = _invoker.ExecuteAsync(
+                    DriverCapability.HistoryRead,
+                    _driver.DriverInstanceId,
+                    async ct => await History.ReadEventsAsync(
+                        sourceName: fullRef,
+                        startUtc: details.StartTime,
+                        endUtc: details.EndTime,
+                        maxEvents: maxEvents,
+                        cancellationToken: ct).ConfigureAwait(false),
+                    CancellationToken.None).AsTask().GetAwaiter().GetResult();
 
                 WriteResult(results, errors, i, StatusCodes.Good,
                     BuildHistoryEvent(driverResult.Events), driverResult.ContinuationPoint);

src/ZB.MOM.WW.OtOpcUa.Server/OpcUa/OpcUaApplicationHost.cs

@@ -3,6 +3,7 @@ using Opc.Ua;
 using Opc.Ua.Configuration;
 using ZB.MOM.WW.OtOpcUa.Core.Hosting;
 using ZB.MOM.WW.OtOpcUa.Core.OpcUa;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
 using ZB.MOM.WW.OtOpcUa.Server.Security;
 
 namespace ZB.MOM.WW.OtOpcUa.Server.OpcUa;
@@ -20,6 +21,7 @@ public sealed class OpcUaApplicationHost : IAsyncDisposable
     private readonly OpcUaServerOptions _options;
     private readonly DriverHost _driverHost;
     private readonly IUserAuthenticator _authenticator;
+    private readonly DriverResiliencePipelineBuilder _pipelineBuilder;
     private readonly ILoggerFactory _loggerFactory;
     private readonly ILogger<OpcUaApplicationHost> _logger;
     private ApplicationInstance? _application;
@@ -27,11 +29,13 @@ public sealed class OpcUaApplicationHost : IAsyncDisposable
     private bool _disposed;
 
     public OpcUaApplicationHost(OpcUaServerOptions options, DriverHost driverHost,
-        IUserAuthenticator authenticator, ILoggerFactory loggerFactory, ILogger<OpcUaApplicationHost> logger)
+        IUserAuthenticator authenticator, ILoggerFactory loggerFactory, ILogger<OpcUaApplicationHost> logger,
+        DriverResiliencePipelineBuilder? pipelineBuilder = null)
     {
         _options = options;
         _driverHost = driverHost;
         _authenticator = authenticator;
+        _pipelineBuilder = pipelineBuilder ?? new DriverResiliencePipelineBuilder();
         _loggerFactory = loggerFactory;
         _logger = logger;
     }
@@ -58,7 +62,7 @@ public sealed class OpcUaApplicationHost : IAsyncDisposable
             throw new InvalidOperationException(
                 $"OPC UA application certificate could not be validated or created in {_options.PkiStoreRoot}");
 
-        _server = new OtOpcUaServer(_driverHost, _authenticator, _loggerFactory);
+        _server = new OtOpcUaServer(_driverHost, _authenticator, _pipelineBuilder, _loggerFactory);
         await _application.Start(_server).ConfigureAwait(false);
 
         _logger.LogInformation("OPC UA server started — endpoint={Endpoint} driverCount={Count}",

src/ZB.MOM.WW.OtOpcUa.Server/OpcUa/OtOpcUaServer.cs

@@ -5,6 +5,7 @@ using Opc.Ua.Server;
 using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
 using ZB.MOM.WW.OtOpcUa.Core.Hosting;
 using ZB.MOM.WW.OtOpcUa.Core.OpcUa;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
 using ZB.MOM.WW.OtOpcUa.Server.Security;
 
 namespace ZB.MOM.WW.OtOpcUa.Server.OpcUa;
@@ -19,13 +20,19 @@ public sealed class OtOpcUaServer : StandardServer
 {
     private readonly DriverHost _driverHost;
     private readonly IUserAuthenticator _authenticator;
+    private readonly DriverResiliencePipelineBuilder _pipelineBuilder;
     private readonly ILoggerFactory _loggerFactory;
     private readonly List<DriverNodeManager> _driverNodeManagers = new();
 
-    public OtOpcUaServer(DriverHost driverHost, IUserAuthenticator authenticator, ILoggerFactory loggerFactory)
+    public OtOpcUaServer(
+        DriverHost driverHost,
+        IUserAuthenticator authenticator,
+        DriverResiliencePipelineBuilder pipelineBuilder,
+        ILoggerFactory loggerFactory)
     {
         _driverHost = driverHost;
         _authenticator = authenticator;
+        _pipelineBuilder = pipelineBuilder;
         _loggerFactory = loggerFactory;
     }
 
@@ -46,7 +53,12 @@ public sealed class OtOpcUaServer : StandardServer
             if (driver is null) continue;
 
             var logger = _loggerFactory.CreateLogger<DriverNodeManager>();
-            var manager = new DriverNodeManager(server, configuration, driver, logger);
+            // Per-driver resilience options: default Tier A pending Stream B.1 which wires
+            // per-type tiers into DriverTypeRegistry. Read ResilienceConfig JSON from the
+            // DriverInstance row in a follow-up PR; for now every driver gets Tier A defaults.
+            var options = new DriverResilienceOptions { Tier = DriverTier.A };
+            var invoker = new CapabilityInvoker(_pipelineBuilder, driver.DriverInstanceId, () => options);
+            var manager = new DriverNodeManager(server, configuration, driver, invoker, logger);
             _driverNodeManagers.Add(manager);
         }
 

tests/ZB.MOM.WW.OtOpcUa.Core.Tests/Resilience/CapabilityInvokerTests.cs

@@ -0,0 +1,151 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Tests.Resilience;
+
+[Trait("Category", "Unit")]
+public sealed class CapabilityInvokerTests
+{
+    private static CapabilityInvoker MakeInvoker(
+        DriverResiliencePipelineBuilder builder,
+        DriverResilienceOptions options) =>
+        new(builder, "drv-test", () => options);
+
+    [Fact]
+    public async Task Read_ReturnsValue_FromCallSite()
+    {
+        var invoker = MakeInvoker(new DriverResiliencePipelineBuilder(), new DriverResilienceOptions { Tier = DriverTier.A });
+
+        var result = await invoker.ExecuteAsync(
+            DriverCapability.Read,
+            "host-1",
+            _ => ValueTask.FromResult(42),
+            CancellationToken.None);
+
+        result.ShouldBe(42);
+    }
+
+    [Fact]
+    public async Task Read_Retries_OnTransientFailure()
+    {
+        var invoker = MakeInvoker(new DriverResiliencePipelineBuilder(), new DriverResilienceOptions { Tier = DriverTier.A });
+        var attempts = 0;
+
+        var result = await invoker.ExecuteAsync(
+            DriverCapability.Read,
+            "host-1",
+            async _ =>
+            {
+                attempts++;
+                if (attempts < 2) throw new InvalidOperationException("transient");
+                await Task.Yield();
+                return "ok";
+            },
+            CancellationToken.None);
+
+        result.ShouldBe("ok");
+        attempts.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task Write_NonIdempotent_DoesNotRetry_EvenWhenPolicyHasRetries()
+    {
+        var options = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Write] = new(TimeoutSeconds: 2, RetryCount: 3, BreakerFailureThreshold: 5),
+            },
+        };
+        var invoker = MakeInvoker(new DriverResiliencePipelineBuilder(), options);
+        var attempts = 0;
+
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await invoker.ExecuteWriteAsync(
+                "host-1",
+                isIdempotent: false,
+                async _ =>
+                {
+                    attempts++;
+                    await Task.Yield();
+                    throw new InvalidOperationException("boom");
+#pragma warning disable CS0162
+                    return 0;
+#pragma warning restore CS0162
+                },
+                CancellationToken.None));
+
+        attempts.ShouldBe(1, "non-idempotent write must never replay");
+    }
+
+    [Fact]
+    public async Task Write_Idempotent_Retries_WhenPolicyHasRetries()
+    {
+        var options = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Write] = new(TimeoutSeconds: 2, RetryCount: 3, BreakerFailureThreshold: 5),
+            },
+        };
+        var invoker = MakeInvoker(new DriverResiliencePipelineBuilder(), options);
+        var attempts = 0;
+
+        var result = await invoker.ExecuteWriteAsync(
+            "host-1",
+            isIdempotent: true,
+            async _ =>
+            {
+                attempts++;
+                if (attempts < 2) throw new InvalidOperationException("transient");
+                await Task.Yield();
+                return "ok";
+            },
+            CancellationToken.None);
+
+        result.ShouldBe("ok");
+        attempts.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task Write_Default_DoesNotRetry_WhenPolicyHasZeroRetries()
+    {
+        // Tier A Write default is RetryCount=0. Even isIdempotent=true shouldn't retry
+        // because the policy says not to.
+        var invoker = MakeInvoker(new DriverResiliencePipelineBuilder(), new DriverResilienceOptions { Tier = DriverTier.A });
+        var attempts = 0;
+
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await invoker.ExecuteWriteAsync(
+                "host-1",
+                isIdempotent: true,
+                async _ =>
+                {
+                    attempts++;
+                    await Task.Yield();
+                    throw new InvalidOperationException("boom");
+#pragma warning disable CS0162
+                    return 0;
+#pragma warning restore CS0162
+                },
+                CancellationToken.None));
+
+        attempts.ShouldBe(1, "tier-A default for Write is RetryCount=0");
+    }
+
+    [Fact]
+    public async Task Execute_HonorsDifferentHosts_Independently()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var invoker = MakeInvoker(builder, new DriverResilienceOptions { Tier = DriverTier.A });
+
+        await invoker.ExecuteAsync(DriverCapability.Read, "host-a", _ => ValueTask.FromResult(1), CancellationToken.None);
+        await invoker.ExecuteAsync(DriverCapability.Read, "host-b", _ => ValueTask.FromResult(2), CancellationToken.None);
+
+        builder.CachedPipelineCount.ShouldBe(2);
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Core.Tests/Resilience/DriverResilienceOptionsTests.cs

@@ -0,0 +1,102 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Tests.Resilience;
+
+[Trait("Category", "Unit")]
+public sealed class DriverResilienceOptionsTests
+{
+    [Theory]
+    [InlineData(DriverTier.A)]
+    [InlineData(DriverTier.B)]
+    [InlineData(DriverTier.C)]
+    public void TierDefaults_Cover_EveryCapability(DriverTier tier)
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(tier);
+
+        foreach (var capability in Enum.GetValues<DriverCapability>())
+            defaults.ShouldContainKey(capability);
+    }
+
+    [Theory]
+    [InlineData(DriverTier.A)]
+    [InlineData(DriverTier.B)]
+    [InlineData(DriverTier.C)]
+    public void Write_NeverRetries_ByDefault(DriverTier tier)
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(tier);
+        defaults[DriverCapability.Write].RetryCount.ShouldBe(0);
+    }
+
+    [Theory]
+    [InlineData(DriverTier.A)]
+    [InlineData(DriverTier.B)]
+    [InlineData(DriverTier.C)]
+    public void AlarmAcknowledge_NeverRetries_ByDefault(DriverTier tier)
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(tier);
+        defaults[DriverCapability.AlarmAcknowledge].RetryCount.ShouldBe(0);
+    }
+
+    [Theory]
+    [InlineData(DriverTier.A, DriverCapability.Read)]
+    [InlineData(DriverTier.A, DriverCapability.HistoryRead)]
+    [InlineData(DriverTier.B, DriverCapability.Discover)]
+    [InlineData(DriverTier.B, DriverCapability.Probe)]
+    [InlineData(DriverTier.C, DriverCapability.AlarmSubscribe)]
+    public void IdempotentCapabilities_Retry_ByDefault(DriverTier tier, DriverCapability capability)
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(tier);
+        defaults[capability].RetryCount.ShouldBeGreaterThan(0);
+    }
+
+    [Fact]
+    public void TierC_DisablesCircuitBreaker_DeferringToSupervisor()
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(DriverTier.C);
+
+        foreach (var (_, policy) in defaults)
+            policy.BreakerFailureThreshold.ShouldBe(0, "Tier C breaker is handled by the Proxy supervisor (decision #68)");
+    }
+
+    [Theory]
+    [InlineData(DriverTier.A)]
+    [InlineData(DriverTier.B)]
+    public void TierAAndB_EnableCircuitBreaker(DriverTier tier)
+    {
+        var defaults = DriverResilienceOptions.GetTierDefaults(tier);
+
+        foreach (var (_, policy) in defaults)
+            policy.BreakerFailureThreshold.ShouldBeGreaterThan(0);
+    }
+
+    [Fact]
+    public void Resolve_Uses_TierDefaults_When_NoOverride()
+    {
+        var options = new DriverResilienceOptions { Tier = DriverTier.A };
+
+        var resolved = options.Resolve(DriverCapability.Read);
+
+        resolved.ShouldBe(DriverResilienceOptions.GetTierDefaults(DriverTier.A)[DriverCapability.Read]);
+    }
+
+    [Fact]
+    public void Resolve_Uses_Override_When_Configured()
+    {
+        var custom = new CapabilityPolicy(TimeoutSeconds: 42, RetryCount: 7, BreakerFailureThreshold: 9);
+        var options = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Read] = custom,
+            },
+        };
+
+        options.Resolve(DriverCapability.Read).ShouldBe(custom);
+        options.Resolve(DriverCapability.Write).ShouldBe(
+            DriverResilienceOptions.GetTierDefaults(DriverTier.A)[DriverCapability.Write]);
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Core.Tests/Resilience/DriverResiliencePipelineBuilderTests.cs

@@ -0,0 +1,222 @@
+using Polly.CircuitBreaker;
+using Polly.Timeout;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Tests.Resilience;
+
+[Trait("Category", "Unit")]
+public sealed class DriverResiliencePipelineBuilderTests
+{
+    private static readonly DriverResilienceOptions TierAOptions = new() { Tier = DriverTier.A };
+
+    [Fact]
+    public async Task Read_Retries_Transient_Failures()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.Read, TierAOptions);
+        var attempts = 0;
+
+        await pipeline.ExecuteAsync(async _ =>
+        {
+            attempts++;
+            if (attempts < 3) throw new InvalidOperationException("transient");
+            await Task.Yield();
+        });
+
+        attempts.ShouldBe(3);
+    }
+
+    [Fact]
+    public async Task Write_DoesNotRetry_OnFailure()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.Write, TierAOptions);
+        var attempts = 0;
+
+        var ex = await Should.ThrowAsync<InvalidOperationException>(async () =>
+        {
+            await pipeline.ExecuteAsync(async _ =>
+            {
+                attempts++;
+                await Task.Yield();
+                throw new InvalidOperationException("boom");
+            });
+        });
+
+        attempts.ShouldBe(1);
+        ex.Message.ShouldBe("boom");
+    }
+
+    [Fact]
+    public async Task AlarmAcknowledge_DoesNotRetry_OnFailure()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.AlarmAcknowledge, TierAOptions);
+        var attempts = 0;
+
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+        {
+            await pipeline.ExecuteAsync(async _ =>
+            {
+                attempts++;
+                await Task.Yield();
+                throw new InvalidOperationException("boom");
+            });
+        });
+
+        attempts.ShouldBe(1);
+    }
+
+    [Fact]
+    public void Pipeline_IsIsolated_PerHost()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var driverId = "drv-test";
+
+        var hostA = builder.GetOrCreate(driverId, "host-a", DriverCapability.Read, TierAOptions);
+        var hostB = builder.GetOrCreate(driverId, "host-b", DriverCapability.Read, TierAOptions);
+
+        hostA.ShouldNotBeSameAs(hostB);
+        builder.CachedPipelineCount.ShouldBe(2);
+    }
+
+    [Fact]
+    public void Pipeline_IsReused_ForSameTriple()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var driverId = "drv-test";
+
+        var first = builder.GetOrCreate(driverId, "host-a", DriverCapability.Read, TierAOptions);
+        var second = builder.GetOrCreate(driverId, "host-a", DriverCapability.Read, TierAOptions);
+
+        first.ShouldBeSameAs(second);
+        builder.CachedPipelineCount.ShouldBe(1);
+    }
+
+    [Fact]
+    public void Pipeline_IsIsolated_PerCapability()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var driverId = "drv-test";
+
+        var read = builder.GetOrCreate(driverId, "host-a", DriverCapability.Read, TierAOptions);
+        var write = builder.GetOrCreate(driverId, "host-a", DriverCapability.Write, TierAOptions);
+
+        read.ShouldNotBeSameAs(write);
+    }
+
+    [Fact]
+    public async Task DeadHost_DoesNotOpenBreaker_ForSiblingHost()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var driverId = "drv-test";
+
+        var deadHost = builder.GetOrCreate(driverId, "dead-plc", DriverCapability.Read, TierAOptions);
+        var liveHost = builder.GetOrCreate(driverId, "live-plc", DriverCapability.Read, TierAOptions);
+
+        var threshold = TierAOptions.Resolve(DriverCapability.Read).BreakerFailureThreshold;
+        for (var i = 0; i < threshold + 5; i++)
+        {
+            await Should.ThrowAsync<Exception>(async () =>
+                await deadHost.ExecuteAsync(async _ =>
+                {
+                    await Task.Yield();
+                    throw new InvalidOperationException("dead plc");
+                }));
+        }
+
+        var liveAttempts = 0;
+        await liveHost.ExecuteAsync(async _ =>
+        {
+            liveAttempts++;
+            await Task.Yield();
+        });
+
+        liveAttempts.ShouldBe(1, "healthy sibling host must not be affected by dead peer");
+    }
+
+    [Fact]
+    public async Task CircuitBreaker_Opens_AfterFailureThreshold_OnTierA()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.Write, TierAOptions);
+
+        var threshold = TierAOptions.Resolve(DriverCapability.Write).BreakerFailureThreshold;
+        for (var i = 0; i < threshold; i++)
+        {
+            await Should.ThrowAsync<InvalidOperationException>(async () =>
+                await pipeline.ExecuteAsync(async _ =>
+                {
+                    await Task.Yield();
+                    throw new InvalidOperationException("boom");
+                }));
+        }
+
+        await Should.ThrowAsync<BrokenCircuitException>(async () =>
+            await pipeline.ExecuteAsync(async _ =>
+            {
+                await Task.Yield();
+            }));
+    }
+
+    [Fact]
+    public async Task Timeout_Cancels_SlowOperation()
+    {
+        var tierAWithShortTimeout = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Read] = new(TimeoutSeconds: 1, RetryCount: 0, BreakerFailureThreshold: 5),
+            },
+        };
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.Read, tierAWithShortTimeout);
+
+        await Should.ThrowAsync<TimeoutRejectedException>(async () =>
+            await pipeline.ExecuteAsync(async ct =>
+            {
+                await Task.Delay(TimeSpan.FromSeconds(5), ct);
+            }));
+    }
+
+    [Fact]
+    public void Invalidate_Removes_OnlyMatchingInstance()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var keepId = "drv-keep";
+        var dropId = "drv-drop";
+
+        builder.GetOrCreate(keepId, "h", DriverCapability.Read,  TierAOptions);
+        builder.GetOrCreate(keepId, "h", DriverCapability.Write, TierAOptions);
+        builder.GetOrCreate(dropId, "h", DriverCapability.Read,  TierAOptions);
+
+        var removed = builder.Invalidate(dropId);
+
+        removed.ShouldBe(1);
+        builder.CachedPipelineCount.ShouldBe(2);
+    }
+
+    [Fact]
+    public async Task Cancellation_IsNot_Retried()
+    {
+        var builder = new DriverResiliencePipelineBuilder();
+        var pipeline = builder.GetOrCreate("drv-test", "host-1", DriverCapability.Read, TierAOptions);
+        var attempts = 0;
+        using var cts = new CancellationTokenSource();
+        cts.Cancel();
+
+        await Should.ThrowAsync<OperationCanceledException>(async () =>
+            await pipeline.ExecuteAsync(async ct =>
+            {
+                attempts++;
+                ct.ThrowIfCancellationRequested();
+                await Task.Yield();
+            }, cts.Token));
+
+        attempts.ShouldBeLessThanOrEqualTo(1);
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Core.Tests/Resilience/FlakeyDriverIntegrationTests.cs

@@ -0,0 +1,160 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Core.Resilience;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Tests.Resilience;
+
+/// <summary>
+///     Integration tests for the Phase 6.1 Stream A.5 contract — wrapping a flaky
+///     <see cref="IReadable"/> / <see cref="IWritable"/> through the <see cref="CapabilityInvoker"/>.
+///     Exercises the three scenarios the plan enumerates: transient read succeeds after N
+///     retries; non-idempotent write fails after one attempt; idempotent write retries through.
+/// </summary>
+[Trait("Category", "Integration")]
+public sealed class FlakeyDriverIntegrationTests
+{
+    [Fact]
+    public async Task Read_SurfacesSuccess_AfterTransientFailures()
+    {
+        var flaky = new FlakeyDriver(failReadsBeforeIndex: 5);
+        var options = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                // TimeoutSeconds=30 gives slack for 5 exponential-backoff retries under
+                // parallel-test-execution CPU pressure; 10 retries at the default Delay=100ms
+                // exponential can otherwise exceed a 2-second budget intermittently.
+                [DriverCapability.Read] = new(TimeoutSeconds: 30, RetryCount: 10, BreakerFailureThreshold: 50),
+            },
+        };
+        var invoker = new CapabilityInvoker(new DriverResiliencePipelineBuilder(), "drv-test", () => options);
+
+        var result = await invoker.ExecuteAsync(
+            DriverCapability.Read,
+            "host-1",
+            async ct => await flaky.ReadAsync(["tag-a"], ct),
+            CancellationToken.None);
+
+        flaky.ReadAttempts.ShouldBe(6);
+        result[0].StatusCode.ShouldBe(0u);
+    }
+
+    [Fact]
+    public async Task Write_NonIdempotent_FailsOnFirstFailure_NoReplay()
+    {
+        var flaky = new FlakeyDriver(failWritesBeforeIndex: 3);
+        var optionsWithAggressiveRetry = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Write] = new(TimeoutSeconds: 2, RetryCount: 5, BreakerFailureThreshold: 50),
+            },
+        };
+        var invoker = new CapabilityInvoker(new DriverResiliencePipelineBuilder(), "drv-test", () => optionsWithAggressiveRetry);
+
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await invoker.ExecuteWriteAsync(
+                "host-1",
+                isIdempotent: false,
+                async ct => await flaky.WriteAsync([new WriteRequest("pulse-coil", true)], ct),
+                CancellationToken.None));
+
+        flaky.WriteAttempts.ShouldBe(1, "non-idempotent write must never replay (decision #44)");
+    }
+
+    [Fact]
+    public async Task Write_Idempotent_RetriesUntilSuccess()
+    {
+        var flaky = new FlakeyDriver(failWritesBeforeIndex: 2);
+        var optionsWithRetry = new DriverResilienceOptions
+        {
+            Tier = DriverTier.A,
+            CapabilityPolicies = new Dictionary<DriverCapability, CapabilityPolicy>
+            {
+                [DriverCapability.Write] = new(TimeoutSeconds: 2, RetryCount: 5, BreakerFailureThreshold: 50),
+            },
+        };
+        var invoker = new CapabilityInvoker(new DriverResiliencePipelineBuilder(), "drv-test", () => optionsWithRetry);
+
+        var results = await invoker.ExecuteWriteAsync(
+            "host-1",
+            isIdempotent: true,
+            async ct => await flaky.WriteAsync([new WriteRequest("set-point", 42.0f)], ct),
+            CancellationToken.None);
+
+        flaky.WriteAttempts.ShouldBe(3);
+        results[0].StatusCode.ShouldBe(0u);
+    }
+
+    [Fact]
+    public async Task MultipleHosts_OnOneDriver_HaveIndependentFailureCounts()
+    {
+        var flaky = new FlakeyDriver(failReadsBeforeIndex: 0);
+        var options = new DriverResilienceOptions { Tier = DriverTier.A };
+        var builder = new DriverResiliencePipelineBuilder();
+        var invoker = new CapabilityInvoker(builder, "drv-test", () => options);
+
+        // host-dead: force many failures to exhaust retries + trip breaker
+        var threshold = options.Resolve(DriverCapability.Read).BreakerFailureThreshold;
+        for (var i = 0; i < threshold + 5; i++)
+        {
+            await Should.ThrowAsync<Exception>(async () =>
+                await invoker.ExecuteAsync(DriverCapability.Read, "host-dead",
+                    _ => throw new InvalidOperationException("dead"),
+                    CancellationToken.None));
+        }
+
+        // host-live: succeeds on first call — unaffected by the dead-host breaker
+        var liveAttempts = 0;
+        await invoker.ExecuteAsync(DriverCapability.Read, "host-live",
+            _ => { liveAttempts++; return ValueTask.FromResult("ok"); },
+            CancellationToken.None);
+
+        liveAttempts.ShouldBe(1);
+    }
+
+    private sealed class FlakeyDriver : IReadable, IWritable
+    {
+        private readonly int _failReadsBeforeIndex;
+        private readonly int _failWritesBeforeIndex;
+
+        public int ReadAttempts { get; private set; }
+        public int WriteAttempts { get; private set; }
+
+        public FlakeyDriver(int failReadsBeforeIndex = 0, int failWritesBeforeIndex = 0)
+        {
+            _failReadsBeforeIndex = failReadsBeforeIndex;
+            _failWritesBeforeIndex = failWritesBeforeIndex;
+        }
+
+        public Task<IReadOnlyList<DataValueSnapshot>> ReadAsync(
+            IReadOnlyList<string> fullReferences,
+            CancellationToken cancellationToken)
+        {
+            var attempt = ++ReadAttempts;
+            if (attempt <= _failReadsBeforeIndex)
+                throw new InvalidOperationException($"transient read failure #{attempt}");
+
+            var now = DateTime.UtcNow;
+            IReadOnlyList<DataValueSnapshot> result = fullReferences
+                .Select(_ => new DataValueSnapshot(Value: 0, StatusCode: 0u, SourceTimestampUtc: now, ServerTimestampUtc: now))
+                .ToList();
+            return Task.FromResult(result);
+        }
+
+        public Task<IReadOnlyList<WriteResult>> WriteAsync(
+            IReadOnlyList<WriteRequest> writes,
+            CancellationToken cancellationToken)
+        {
+            var attempt = ++WriteAttempts;
+            if (attempt <= _failWritesBeforeIndex)
+                throw new InvalidOperationException($"transient write failure #{attempt}");
+
+            IReadOnlyList<WriteResult> result = writes.Select(_ => new WriteResult(0u)).ToList();
+            return Task.FromResult(result);
+        }
+    }
+}

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

@@ -220,6 +220,23 @@ public sealed class ModbusDriverTests
         builder.Variables.ShouldContain(v => v.BrowseName == "Run" && v.Info.DriverDataType == DriverDataType.Boolean);
     }
 
+    [Fact]
+    public async Task Discover_propagates_WriteIdempotent_from_tag_to_attribute_info()
+    {
+        var (drv, _) = NewDriver(
+            new ModbusTagDefinition("SetPoint", ModbusRegion.HoldingRegisters, 0, ModbusDataType.Float32, WriteIdempotent: true),
+            new ModbusTagDefinition("PulseCoil", ModbusRegion.Coils, 0, ModbusDataType.Bool));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var builder = new RecordingBuilder();
+        await drv.DiscoverAsync(builder, CancellationToken.None);
+
+        var setPoint = builder.Variables.Single(v => v.BrowseName == "SetPoint");
+        var pulse = builder.Variables.Single(v => v.BrowseName == "PulseCoil");
+        setPoint.Info.WriteIdempotent.ShouldBeTrue();
+        pulse.Info.WriteIdempotent.ShouldBeFalse("default is opt-in per decision #44");
+    }
+
     // --- helpers ---
 
     private sealed class RecordingBuilder : IAddressSpaceBuilder

tests/ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient.Tests/OpcUaClientAlarmTests.cs

@@ -0,0 +1,70 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient.Tests;
+
+[Trait("Category", "Unit")]
+public sealed class OpcUaClientAlarmTests
+{
+    [Theory]
+    [InlineData((ushort)1, AlarmSeverity.Low)]
+    [InlineData((ushort)200, AlarmSeverity.Low)]
+    [InlineData((ushort)201, AlarmSeverity.Medium)]
+    [InlineData((ushort)500, AlarmSeverity.Medium)]
+    [InlineData((ushort)501, AlarmSeverity.High)]
+    [InlineData((ushort)800, AlarmSeverity.High)]
+    [InlineData((ushort)801, AlarmSeverity.Critical)]
+    [InlineData((ushort)1000, AlarmSeverity.Critical)]
+    public void MapSeverity_buckets_per_OPC_UA_Part_9_guidance(ushort opcSev, AlarmSeverity expected)
+    {
+        OpcUaClientDriver.MapSeverity(opcSev).ShouldBe(expected);
+    }
+
+    [Fact]
+    public void MapSeverity_zero_maps_to_Low()
+    {
+        // 0 isn't in OPC UA's 1-1000 range but we handle it gracefully as Low.
+        OpcUaClientDriver.MapSeverity(0).ShouldBe(AlarmSeverity.Low);
+    }
+
+    [Fact]
+    public async Task SubscribeAlarmsAsync_without_initialize_throws_InvalidOperationException()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-alarm-uninit");
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await drv.SubscribeAlarmsAsync([], TestContext.Current.CancellationToken));
+    }
+
+    [Fact]
+    public async Task UnsubscribeAlarmsAsync_with_unknown_handle_is_noop()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-alarm-unknown");
+        // Parallels the subscribe handle path — session-drop races shouldn't crash the caller.
+        await drv.UnsubscribeAlarmsAsync(new FakeAlarmHandle(), TestContext.Current.CancellationToken);
+    }
+
+    [Fact]
+    public async Task AcknowledgeAsync_without_initialize_throws_InvalidOperationException()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-ack-uninit");
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await drv.AcknowledgeAsync(
+                [new AlarmAcknowledgeRequest("ns=2;s=Src", "ns=2;s=Cond", "operator ack")],
+                TestContext.Current.CancellationToken));
+    }
+
+    [Fact]
+    public async Task AcknowledgeAsync_with_empty_batch_is_noop_even_without_init()
+    {
+        // Empty batch short-circuits before touching the session, so it's safe pre-init. This
+        // keeps batch-ack callers from needing to guard the list size themselves.
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-ack-empty");
+        await drv.AcknowledgeAsync([], TestContext.Current.CancellationToken);
+    }
+
+    private sealed class FakeAlarmHandle : IAlarmSubscriptionHandle
+    {
+        public string DiagnosticId => "fake-alarm";
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient.Tests/OpcUaClientHistoryTests.cs

@@ -0,0 +1,91 @@
+using Opc.Ua;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.OpcUaClient.Tests;
+
+[Trait("Category", "Unit")]
+public sealed class OpcUaClientHistoryTests
+{
+    [Theory]
+    [InlineData(HistoryAggregateType.Average)]
+    [InlineData(HistoryAggregateType.Minimum)]
+    [InlineData(HistoryAggregateType.Maximum)]
+    [InlineData(HistoryAggregateType.Total)]
+    [InlineData(HistoryAggregateType.Count)]
+    public void MapAggregateToNodeId_returns_standard_Part13_aggregate_for_every_enum(HistoryAggregateType agg)
+    {
+        var nodeId = OpcUaClientDriver.MapAggregateToNodeId(agg);
+        NodeId.IsNull(nodeId).ShouldBeFalse();
+        // Every mapping should resolve to an AggregateFunction_* NodeId (namespace 0, numeric id).
+        nodeId.NamespaceIndex.ShouldBe((ushort)0);
+    }
+
+    [Fact]
+    public void MapAggregateToNodeId_rejects_invalid_enum_value()
+    {
+        // Defense-in-depth: a future HistoryAggregateType addition mustn't silently fall through.
+        Should.Throw<ArgumentOutOfRangeException>(() =>
+            OpcUaClientDriver.MapAggregateToNodeId((HistoryAggregateType)99));
+    }
+
+    [Fact]
+    public async Task ReadRawAsync_without_initialize_throws_InvalidOperationException()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-hist-uninit");
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await drv.ReadRawAsync("ns=2;s=Counter",
+                DateTime.UtcNow.AddMinutes(-5), DateTime.UtcNow, 1000,
+                TestContext.Current.CancellationToken));
+    }
+
+    [Fact]
+    public async Task ReadRawAsync_with_malformed_NodeId_returns_empty_result_not_throw()
+    {
+        // Same defensive pattern as ReadAsync / WriteAsync — malformed NodeId short-circuits
+        // to an empty result rather than crashing a batch history call. Needs init via the
+        // throw path first, then we pass "" to trigger the parse-fail branch inside
+        // ExecuteHistoryReadAsync. The init itself fails against 127.0.0.1:1 so we stop there.
+        // Not runnable without init — keep as placeholder for when the in-process fixture
+        // PR lands.
+        await Task.CompletedTask;
+    }
+
+    [Fact]
+    public async Task ReadProcessedAsync_without_initialize_throws_InvalidOperationException()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-hist-uninit");
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await drv.ReadProcessedAsync("ns=2;s=Counter",
+                DateTime.UtcNow.AddMinutes(-5), DateTime.UtcNow,
+                TimeSpan.FromSeconds(10), HistoryAggregateType.Average,
+                TestContext.Current.CancellationToken));
+    }
+
+    [Fact]
+    public async Task ReadAtTimeAsync_without_initialize_throws_InvalidOperationException()
+    {
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-hist-uninit");
+        await Should.ThrowAsync<InvalidOperationException>(async () =>
+            await drv.ReadAtTimeAsync("ns=2;s=Counter",
+                [DateTime.UtcNow.AddMinutes(-5), DateTime.UtcNow],
+                TestContext.Current.CancellationToken));
+    }
+
+    [Fact]
+    public async Task ReadEventsAsync_throws_NotSupportedException_as_documented()
+    {
+        // The IHistoryProvider default implementation throws; the OPC UA Client driver
+        // deliberately inherits that default (see PR 76 commit body) because the OPC UA
+        // client call path needs an EventFilter SelectClauses spec the interface doesn't carry.
+        using var drv = new OpcUaClientDriver(new OpcUaClientDriverOptions(), "opcua-events-default");
+        await Should.ThrowAsync<NotSupportedException>(async () =>
+            await ((IHistoryProvider)drv).ReadEventsAsync(
+                sourceName: null,
+                startUtc: DateTime.UtcNow.AddMinutes(-5),
+                endUtc: DateTime.UtcNow,
+                maxEvents: 100,
+                cancellationToken: TestContext.Current.CancellationToken));
+    }
+}

tests/ZB.MOM.WW.OtOpcUa.Driver.S7.Tests/S7DiscoveryAndSubscribeTests.cs

@@ -65,6 +65,27 @@ public sealed class S7DiscoveryAndSubscribeTests
         builder.Variables[2].Attr.DriverDataType.ShouldBe(DriverDataType.Float32);
     }
 
+    [Fact]
+    public async Task DiscoverAsync_propagates_WriteIdempotent_from_tag_to_attribute_info()
+    {
+        var opts = new S7DriverOptions
+        {
+            Host = "192.0.2.1",
+            Tags =
+            [
+                new("SetPoint", "DB1.DBW0", S7DataType.Int16, WriteIdempotent: true),
+                new("StartBit", "M0.0", S7DataType.Bool),
+            ],
+        };
+        using var drv = new S7Driver(opts, "s7-idem");
+
+        var builder = new RecordingAddressSpaceBuilder();
+        await drv.DiscoverAsync(builder, TestContext.Current.CancellationToken);
+
+        builder.Variables.Single(v => v.Name == "SetPoint").Attr.WriteIdempotent.ShouldBeTrue();
+        builder.Variables.Single(v => v.Name == "StartBit").Attr.WriteIdempotent.ShouldBeFalse("default is opt-in per decision #44");
+    }
+
     [Fact]
     public void GetHostStatuses_returns_one_row_with_host_port_identity_pre_init()
     {