feat(transport): fleet-wide bundle-unlock throttle window (register row 52)

The per-IP unlock throttle's counters were a process-local ConcurrentDictionary
on each central node, so the effective fleet cap was N x
MaxUnlockAttemptsPerIpPerHour per client (failing over to the other node handed
the client a fresh budget) and a node restart cleared the window outright.

The window now lives in the central MS SQL BundleUnlockAttempt table — the HA
store both central nodes already share, so no new infrastructure, no new port,
no consensus protocol:

- Commons: BundleUnlockAttempt entity + IBundleUnlockAttemptRepository.
- ConfigurationDatabase: entity config (one index,
  IX_BundleUnlockAttempt_Client_Attempted, serving both the windowed count and
  the prune), repository (AsNoTracking count, ExecuteDeleteAsync prune),
  DbSet, DI registration, migration AddBundleUnlockAttemptTable + the
  hand-written idempotent production script under docs/plans/sql/.
- Transport: IBundleUnlockAttemptWindow seam + SqlBundleUnlockAttemptWindow
  (singleton; resolves the scoped repository from its own scope per call, like
  the central purge singletons). BundleUnlockRateLimiter keeps the policy and
  gains TryRegisterAttemptAsync; BundleImporter.LoadAsync calls it.

Behaviour preserved from row 44: locked bundles spend no hourly budget (the
lockout check still runs first), throttle denials burn no per-bundle strike and
are NOT recorded so they cannot roll the window forward, unattributable clients
share the stricter <unattributed> bucket, and the budget counts attempts rather
than failures. Config keys and their semantics are unchanged, so the rig's 500
override stays valid.

Three deliberate trades, documented in both Transport docs:
- FAILS OPEN. A store error logs a Warning naming the client and cap, and
  permits the attempt. This is a best-effort rate cap over an already
  authenticated, Administrator-gated surface — the availability of legitimate
  imports outranks the cap, and a DB blip must not make the import wizard
  unusable. Falling back to the process-local window was considered and
  rejected: it would still block a legitimate operator's 11th import.
- Approximate. Count-then-insert is not atomic across nodes, so simultaneous
  attempts can overshoot by the number in flight — bounded, self-correcting
  within the window, and not worth provider-specific raw SQL.
- Pruned opportunistically on the unlock path (CAS-claimed at most every 15
  min per process, bounded to 1000 rows, retaining 2x the window, best-effort
  try/catch) rather than by a new background service — Transport owns no
  central scheduler. Same shape as DeploymentService's terminal-record sweep.

Tests: Transport.IntegrationTests 116 passed (+10 — two limiter instances over
one store enforcing a single combined window, restart durability, denials
burning no budget, sliding release, bucket isolation, store-failure fail-open,
caller-cancellation not swallowed, no-shared-window fallback, prune and its
negative control); ConfigurationDatabase 395 passed (+5).

src/ZB.MOM.WW.ScadaBridge.Transport/Import/BundleImporter.cs

@@ -293,11 +293,18 @@ public sealed class BundleImporter : IBundleImporter
             //     operator is legitimately importing.
             // Registered on every unlock attempt (successful ones included) because the
             // documented budget counts attempts, not failures.
+            //
+            // Register row 52: the async overload consults the FLEET-WIDE window (central
+            // MS SQL) so the cap is not multiplied by the number of central nodes and does
+            // not reset on a node restart. It fails open on a store error — see
+            // BundleUnlockRateLimiter's remarks.
             var throttleKey = string.IsNullOrWhiteSpace(clientKey)
                 ? UnattributedClientKey
                 : clientKey;
             var perIpCap = _options.Value.MaxUnlockAttemptsPerIpPerHour;
-            if (!_unlockRateLimiter.TryRegisterAttempt(throttleKey, perIpCap))
+            if (!await _unlockRateLimiter
+                    .TryRegisterAttemptAsync(throttleKey, perIpCap, ct)
+                    .ConfigureAwait(false))
             {
                 _logger?.LogWarning(
                     "Bundle unlock throttled: client {ClientKey} exceeded {Cap} attempts in the trailing hour.",

src/ZB.MOM.WW.ScadaBridge.Transport/Import/BundleUnlockRateLimiter.cs

@@ -1,26 +1,37 @@
 using System.Collections.Concurrent;
+using Microsoft.Extensions.Logging;
+using Microsoft.Extensions.Logging.Abstractions;
 
 namespace ZB.MOM.WW.ScadaBridge.Transport.Import;
 
 /// <summary>
-/// In-memory sliding-window rate limiter for bundle-unlock passphrase
-/// attempts, keyed by client IP. The design doc (§11) declares a per-IP-per-hour cap
-/// (default 10) as a brute-force defence against a stolen bundle; this class is the
-/// minimal server-side implementation.
+/// Sliding-window rate limiter for bundle-unlock passphrase attempts, keyed by client IP.
+/// The design doc (§11) declares a per-IP-per-hour cap (default 10) as a brute-force
+/// defence against a stolen bundle; this class owns that policy.
 /// <para>
-/// Algorithm: each key (an IP string, or any opaque caller identifier) holds a queue
-/// of attempt timestamps. <see cref="TryRegisterAttempt"/> first prunes entries older
-/// than the configured window, then either appends the current timestamp and returns
-/// <c>true</c> if the count is still under the threshold, or refuses to append and
-/// returns <c>false</c> if appending would cross it. The trailing-hour count is the
-/// queue length post-prune.
+/// <b>The window is fleet-wide when an <see cref="IBundleUnlockAttemptWindow"/> is
+/// supplied (register row 52).</b> The counters used to be per-node AND per-process, so
+/// the effective cap was N x limit per client across the central pair and a node restart
+/// silently cleared the window. <see cref="TryRegisterAttemptAsync"/> now consults the
+/// shared central-MS-SQL window instead, which both central nodes already share. The
+/// production composition root always supplies one.
 /// </para>
 /// <para>
-/// Storage is a process-local <see cref="ConcurrentDictionary{TKey,TValue}"/>. The
-/// counters do not survive a host restart — that is by design: a restart resets the
-/// brute-force window in favour of legitimate operators after an outage. Persisting
-/// the counters would require a multi-node consensus story the simple in-memory
-/// design avoids.
+/// <b>Fail-open is deliberate.</b> A store failure in the throttle path logs a Warning and
+/// PERMITS the attempt. This is a best-effort rate cap layered over a surface that is
+/// already authenticated and role-gated (Administrator-only bundle import), so the
+/// availability of legitimate imports outranks the cap: a database blip must not make the
+/// import wizard unusable. Falling back to the process-local window instead was considered
+/// and rejected — it would still block a legitimate operator's 11th import during an
+/// outage, which is the outcome fail-open exists to prevent.
+/// </para>
+/// <para>
+/// Algorithm (the process-local fallback used when no shared window is registered): each
+/// key holds a queue of attempt timestamps. <see cref="TryRegisterAttempt"/> first prunes
+/// entries older than the configured window, then either appends the current timestamp and
+/// returns <c>true</c> if the count is still under the threshold, or refuses to append and
+/// returns <c>false</c> if appending would cross it. Either way a REJECTED attempt is never
+/// appended, so a denial burns no budget — the same contract the shared window implements.
 /// </para>
 /// <para>
 /// Thread-safety: the per-key queue is protected by a per-key lock taken inside the
@@ -39,10 +50,13 @@ public sealed class BundleUnlockRateLimiter
     private readonly ConcurrentDictionary<string, AttemptBucket> _buckets = new(StringComparer.OrdinalIgnoreCase);
     private readonly TimeProvider _timeProvider;
     private readonly TimeSpan _window;
+    private readonly IBundleUnlockAttemptWindow? _sharedWindow;
+    private readonly ILogger<BundleUnlockRateLimiter> _logger;
 
     /// <summary>
     /// Initializes a new <see cref="BundleUnlockRateLimiter"/> using the documented
-    /// 1-hour trailing window and the system clock. Suitable for production DI.
+    /// 1-hour trailing window and the system clock, with no shared window (process-local
+    /// counters only).
     /// </summary>
     public BundleUnlockRateLimiter() : this(TimeProvider.System, DefaultWindow)
     {
@@ -50,11 +64,28 @@ public sealed class BundleUnlockRateLimiter
 
     /// <summary>
     /// Initializes a new <see cref="BundleUnlockRateLimiter"/> with an injected clock
-    /// (for deterministic tests) and a custom trailing window.
+    /// (for deterministic tests) and a custom trailing window, with no shared window.
     /// </summary>
     /// <param name="timeProvider">Clock used for both timestamping new attempts and pruning expired ones.</param>
     /// <param name="window">Trailing window over which attempts are counted (typically 1 hour).</param>
     public BundleUnlockRateLimiter(TimeProvider timeProvider, TimeSpan window)
+        : this(timeProvider, window, sharedWindow: null, logger: null)
+    {
+    }
+
+    /// <summary>
+    /// Initializes a new <see cref="BundleUnlockRateLimiter"/> over a fleet-wide shared
+    /// window (register row 52). This is the overload the production composition root uses.
+    /// </summary>
+    /// <param name="timeProvider">Clock used for the process-local fallback window.</param>
+    /// <param name="window">Trailing window over which attempts are counted (typically 1 hour).</param>
+    /// <param name="sharedWindow">Shared cross-node attempt window; when null the limiter falls back to process-local counters.</param>
+    /// <param name="logger">Optional logger. Used for the fail-open Warning when the shared window is unavailable.</param>
+    public BundleUnlockRateLimiter(
+        TimeProvider timeProvider,
+        TimeSpan window,
+        IBundleUnlockAttemptWindow? sharedWindow,
+        ILogger<BundleUnlockRateLimiter>? logger)
     {
         ArgumentNullException.ThrowIfNull(timeProvider);
         if (window <= TimeSpan.Zero)
@@ -64,13 +95,81 @@ public sealed class BundleUnlockRateLimiter
 
         _timeProvider = timeProvider;
         _window = window;
+        _sharedWindow = sharedWindow;
+        _logger = logger ?? NullLogger<BundleUnlockRateLimiter>.Instance;
     }
 
     /// <summary>
-    /// Attempts to register a new passphrase try against the configured per-key
-    /// limit. Returns <c>true</c> when the attempt is permitted (and recorded);
-    /// returns <c>false</c> when the key has exhausted its budget for the trailing
-    /// window — the caller should reject the unlock request with a 429-equivalent.
+    /// Gets a value indicating whether this limiter is backed by a fleet-wide shared
+    /// window. False means the cap degrades to per-node, per-process counters.
+    /// </summary>
+    public bool IsFleetWide => _sharedWindow is not null;
+
+    /// <summary>
+    /// Registers a passphrase try against the configured per-key limit, using the
+    /// fleet-wide shared window when one is registered and the process-local window
+    /// otherwise. This is the method the importer calls.
+    /// </summary>
+    /// <param name="clientKey">Opaque caller identifier — typically the remote IP.</param>
+    /// <param name="maxAttemptsPerWindow">The trailing-window cap (<c>MaxUnlockAttemptsPerIpPerHour</c>).</param>
+    /// <param name="ct">Cancellation token.</param>
+    /// <returns>
+    /// <c>true</c> if the attempt is permitted; <c>false</c> if the caller has exhausted
+    /// its budget. Also <c>true</c> when the shared window is unavailable — see the
+    /// class remarks on why the throttle fails open.
+    /// </returns>
+    public async Task<bool> TryRegisterAttemptAsync(
+        string clientKey, int maxAttemptsPerWindow, CancellationToken ct = default)
+    {
+        ArgumentException.ThrowIfNullOrWhiteSpace(clientKey);
+        if (maxAttemptsPerWindow < 1)
+        {
+            throw new ArgumentOutOfRangeException(
+                nameof(maxAttemptsPerWindow), "Limit must be at least 1.");
+        }
+
+        if (_sharedWindow is null)
+        {
+            return TryRegisterAttempt(clientKey, maxAttemptsPerWindow);
+        }
+
+        try
+        {
+            return await _sharedWindow
+                .TryRegisterAttemptAsync(clientKey, maxAttemptsPerWindow, _window, ct)
+                .ConfigureAwait(false);
+        }
+        catch (OperationCanceledException) when (ct.IsCancellationRequested)
+        {
+            // The caller's own cancellation, not a store failure — do not swallow it.
+            throw;
+        }
+        catch (Exception ex)
+        {
+            // Fail OPEN: a best-effort rate cap must never take the import surface down
+            // with the database. The Warning is the operator's signal that the fleet cap
+            // is currently not being enforced.
+            _logger.LogWarning(
+                ex,
+                "Fleet-wide bundle-unlock throttle unavailable; permitting the unlock attempt for client "
+                + "{ClientKey} without enforcing the {Cap}/hour cap.",
+                clientKey, maxAttemptsPerWindow);
+            return true;
+        }
+    }
+
+    /// <summary>
+    /// Attempts to register a new passphrase try against the configured per-key limit in
+    /// the PROCESS-LOCAL window. Returns <c>true</c> when the attempt is permitted (and
+    /// recorded); returns <c>false</c> when the key has exhausted its budget for the
+    /// trailing window — the caller should reject the unlock request with a 429-equivalent.
+    /// <para>
+    /// Register row 52: this bypasses the fleet-wide window. Callers on the unlock path
+    /// must use <see cref="TryRegisterAttemptAsync"/>, which routes to the shared window
+    /// when one is registered and falls back here otherwise. This overload remains public
+    /// for hosts that register Transport without a shared window, and for the tests that
+    /// pin the sliding-window arithmetic directly.
+    /// </para>
     /// </summary>
     /// <param name="clientKey">
     /// Opaque caller identifier — typically the remote IP, but any stable per-source

src/ZB.MOM.WW.ScadaBridge.Transport/Import/IBundleUnlockAttemptWindow.cs

@@ -0,0 +1,28 @@
+namespace ZB.MOM.WW.ScadaBridge.Transport.Import;
+
+/// <summary>
+/// Register row 52: a sliding window of bundle-unlock attempts that is shared across the
+/// central pair, so <c>TransportOptions.MaxUnlockAttemptsPerIpPerHour</c> is a FLEET cap
+/// rather than a per-node, per-process one.
+/// <para>
+/// Implementations are expected to THROW on store failure. The fail-open decision belongs
+/// to <see cref="BundleUnlockRateLimiter"/>, which owns the policy (log a Warning and
+/// permit the attempt), not to the storage layer.
+/// </para>
+/// </summary>
+public interface IBundleUnlockAttemptWindow
+{
+    /// <summary>
+    /// Registers one unlock attempt against the shared window and reports whether it was
+    /// within budget. Implementations must NOT record an attempt they are rejecting —
+    /// a denial burns no budget, matching the in-memory limiter's contract (register
+    /// row 44), so a client that keeps hammering cannot extend its own lockout.
+    /// </summary>
+    /// <param name="clientKey">Throttle bucket — the caller's remote IP, or the shared unattributed bucket.</param>
+    /// <param name="maxAttemptsPerWindow">The trailing-window cap (<c>MaxUnlockAttemptsPerIpPerHour</c>).</param>
+    /// <param name="window">Trailing window the attempts are counted over (one hour in production).</param>
+    /// <param name="ct">Cancellation token.</param>
+    /// <returns><c>true</c> when the attempt was recorded (within budget); <c>false</c> when the bucket is exhausted.</returns>
+    Task<bool> TryRegisterAttemptAsync(
+        string clientKey, int maxAttemptsPerWindow, TimeSpan window, CancellationToken ct = default);
+}

src/ZB.MOM.WW.ScadaBridge.Transport/Import/SqlBundleUnlockAttemptWindow.cs

@@ -0,0 +1,167 @@
+using Microsoft.Extensions.DependencyInjection;
+using Microsoft.Extensions.Logging;
+using Microsoft.Extensions.Logging.Abstractions;
+using ZB.MOM.WW.ScadaBridge.Commons.Interfaces.Repositories;
+
+namespace ZB.MOM.WW.ScadaBridge.Transport.Import;
+
+/// <summary>
+/// Register row 52: <see cref="IBundleUnlockAttemptWindow"/> backed by the central MS SQL
+/// <c>BundleUnlockAttempt</c> table — the HA store both central nodes already share, so
+/// the fleet-wide cap needs no new infrastructure, no new port and no consensus protocol.
+/// <para>
+/// Registered as a singleton (the limiter that consumes it is one) and therefore resolves
+/// a scoped <see cref="IBundleUnlockAttemptRepository"/> from its own
+/// <see cref="IServiceScopeFactory"/> scope per call, the same way the central purge
+/// singletons reach the database.
+/// </para>
+/// <para>
+/// <b>Accuracy is deliberately approximate.</b> The count-then-insert pair is not atomic
+/// across the two central nodes, so genuinely simultaneous attempts can overshoot the cap
+/// by up to the number in flight. A conditional single-statement insert would close that,
+/// at the cost of provider-specific SQL for a best-effort rate cap layered over an
+/// already-authenticated, role-gated surface. Overshoot is bounded and self-corrects
+/// within the window; it is not worth raw SQL.
+/// </para>
+/// </summary>
+public sealed class SqlBundleUnlockAttemptWindow : IBundleUnlockAttemptWindow
+{
+    /// <summary>
+    /// Minimum wall-clock gap between opportunistic prunes. The prune runs on the unlock
+    /// path (Transport owns no central scheduler, and the design constraint was "no new
+    /// background service"), so it is claimed at most this often per process — the same
+    /// rate-limited opportunistic-sweep shape <c>DeploymentService</c> uses for terminal
+    /// deployment records.
+    /// </summary>
+    internal static readonly TimeSpan PruneMinInterval = TimeSpan.FromMinutes(15);
+
+    /// <summary>
+    /// Upper bound on rows deleted by one opportunistic prune pass. Bounded so a sweep on
+    /// a request path can never become a long-running delete, even if the table grew while
+    /// prunes were failing.
+    /// </summary>
+    internal const int PruneBatchSize = 1000;
+
+    /// <summary>
+    /// Multiple of the trailing window that rows are retained for before becoming
+    /// prunable. Kept above 1 so a prune racing a count can never delete a row the count
+    /// still needs.
+    /// </summary>
+    internal const int PruneRetentionWindows = 2;
+
+    private readonly IServiceScopeFactory _scopeFactory;
+    private readonly TimeProvider _timeProvider;
+    private readonly ILogger<SqlBundleUnlockAttemptWindow> _logger;
+
+    /// <summary>Ticks of the last claimed prune, as a CAS-guarded process-local claim.</summary>
+    private long _lastPruneTicks;
+
+    /// <summary>
+    /// Initializes a new <see cref="SqlBundleUnlockAttemptWindow"/>.
+    /// </summary>
+    /// <param name="scopeFactory">Scope factory used to resolve the scoped attempt repository per call.</param>
+    /// <param name="timeProvider">Clock used for the window boundary and the prune cadence.</param>
+    /// <param name="logger">Optional logger for prune diagnostics.</param>
+    public SqlBundleUnlockAttemptWindow(
+        IServiceScopeFactory scopeFactory,
+        TimeProvider timeProvider,
+        ILogger<SqlBundleUnlockAttemptWindow>? logger = null)
+    {
+        _scopeFactory = scopeFactory ?? throw new ArgumentNullException(nameof(scopeFactory));
+        _timeProvider = timeProvider ?? throw new ArgumentNullException(nameof(timeProvider));
+        _logger = logger ?? NullLogger<SqlBundleUnlockAttemptWindow>.Instance;
+
+        // Seed the prune claim to "now" so a freshly started node does not prune on its
+        // very first unlock — the table is almost certainly already tidy, and a node that
+        // restarts under load would otherwise prune on every restart.
+        _lastPruneTicks = _timeProvider.GetUtcNow().UtcTicks;
+    }
+
+    /// <inheritdoc />
+    public async Task<bool> TryRegisterAttemptAsync(
+        string clientKey, int maxAttemptsPerWindow, TimeSpan window, CancellationToken ct = default)
+    {
+        ArgumentException.ThrowIfNullOrWhiteSpace(clientKey);
+        if (maxAttemptsPerWindow < 1)
+        {
+            throw new ArgumentOutOfRangeException(
+                nameof(maxAttemptsPerWindow), "Limit must be at least 1.");
+        }
+
+        if (window <= TimeSpan.Zero)
+        {
+            throw new ArgumentOutOfRangeException(nameof(window), "Window must be positive.");
+        }
+
+        var now = _timeProvider.GetUtcNow().UtcDateTime;
+        var key = clientKey.Trim();
+
+        using var scope = _scopeFactory.CreateScope();
+        var repo = scope.ServiceProvider.GetRequiredService<IBundleUnlockAttemptRepository>();
+
+        // Two round trips rather than one conditional insert — see the class remarks on
+        // why the resulting overshoot is accepted. Exceptions propagate: the fail-open
+        // decision lives in BundleUnlockRateLimiter.
+        var used = await repo.CountAttemptsSinceAsync(key, now - window, ct).ConfigureAwait(false);
+        if (used >= maxAttemptsPerWindow)
+        {
+            // Rejected attempts are NOT recorded — a denial burns no budget, so a client
+            // hammering the endpoint cannot roll its own window forward indefinitely.
+            return false;
+        }
+
+        await repo.RecordAttemptAsync(key, now, ct).ConfigureAwait(false);
+
+        await TryPruneAsync(repo, now, window, ct).ConfigureAwait(false);
+        return true;
+    }
+
+    /// <summary>
+    /// Opportunistic, rate-limited, bounded prune of aged-out attempt rows. Best-effort:
+    /// any failure is logged and swallowed, because retention maintenance must never fail
+    /// the operator's actual unlock.
+    /// </summary>
+    /// <param name="repo">Repository resolved for the current call's scope.</param>
+    /// <param name="nowUtc">Current UTC instant.</param>
+    /// <param name="window">The trailing window; rows older than a multiple of it are prunable.</param>
+    /// <param name="ct">Cancellation token.</param>
+    /// <returns>A task that completes when the prune attempt has finished (or been skipped).</returns>
+    private async Task TryPruneAsync(
+        IBundleUnlockAttemptRepository repo, DateTime nowUtc, TimeSpan window, CancellationToken ct)
+    {
+        var previous = Interlocked.Read(ref _lastPruneTicks);
+        if (nowUtc.Ticks - previous < PruneMinInterval.Ticks)
+        {
+            return;
+        }
+
+        // Claim the sweep so concurrent unlocks on this node don't all prune at once.
+        if (Interlocked.CompareExchange(ref _lastPruneTicks, nowUtc.Ticks, previous) != previous)
+        {
+            return;
+        }
+
+        try
+        {
+            var cutoff = nowUtc - (window * PruneRetentionWindows);
+            var deleted = await repo
+                .PruneAttemptsOlderThanAsync(cutoff, PruneBatchSize, ct)
+                .ConfigureAwait(false);
+
+            if (deleted > 0)
+            {
+                _logger.LogDebug(
+                    "Pruned {Deleted} bundle-unlock attempt row(s) older than {Cutoff:o}.", deleted, cutoff);
+            }
+        }
+        catch (OperationCanceledException) when (ct.IsCancellationRequested)
+        {
+            throw;
+        }
+        catch (Exception ex)
+        {
+            // Never fail the unlock for a retention sweep.
+            _logger.LogWarning(ex, "Opportunistic bundle-unlock attempt prune failed; will retry on a later attempt.");
+        }
+    }
+}

src/ZB.MOM.WW.ScadaBridge.Transport/ServiceCollectionExtensions.cs

@@ -1,5 +1,6 @@
 using Microsoft.Extensions.DependencyInjection;
 using Microsoft.Extensions.DependencyInjection.Extensions;
+using Microsoft.Extensions.Logging;
 using Microsoft.Extensions.Options;
 using ZB.MOM.WW.ScadaBridge.Commons.Interfaces.Transport;
 using ZB.MOM.WW.ScadaBridge.TemplateEngine.Validation;
@@ -43,10 +44,22 @@ public static class ServiceCollectionExtensions
         services.AddScoped<DependencyResolver>();
         services.AddScoped<IBundleExporter, BundleExporter>();
         services.AddSingleton<IBundleSessionStore, BundleSessionStore>();
-        // T-004: per-IP-per-hour unlock rate limiter — design doc §11. Singleton
-        // so the trailing-hour window is shared across every importer scope; the
-        // counters live in-memory and reset on host restart (by design).
-        services.AddSingleton<BundleUnlockRateLimiter>();
+        // T-004: per-IP-per-hour unlock rate limiter — design doc §11. Singleton so the
+        // trailing-hour window is shared across every importer scope.
+        //
+        // Register row 52: the window is FLEET-WIDE, backed by the central MS SQL
+        // BundleUnlockAttempt table that both central nodes already share — the counters
+        // used to be per-node and per-process, so the effective cap was N x limit per
+        // client and a restart cleared the window. The window resolves its scoped
+        // repository from its own scope per call, so it is safe as a singleton.
+        // Constructed explicitly rather than by convention so it is unambiguous which
+        // constructor overload production uses.
+        services.AddSingleton<IBundleUnlockAttemptWindow, SqlBundleUnlockAttemptWindow>();
+        services.AddSingleton(sp => new BundleUnlockRateLimiter(
+            sp.GetRequiredService<TimeProvider>(),
+            BundleUnlockRateLimiter.DefaultWindow,
+            sp.GetRequiredService<IBundleUnlockAttemptWindow>(),
+            sp.GetService<ILogger<BundleUnlockRateLimiter>>()));
         // T-007: periodic eviction sweep so abandoned sessions clear without
         // needing a fresh Get() to trigger lazy eviction.
         services.AddHostedService<BundleSessionEvictionService>();