diff --git a/docs/Historian.md b/docs/Historian.md
index 97667113..78a29031 100644
--- a/docs/Historian.md
+++ b/docs/Historian.md
@@ -587,14 +587,20 @@ paging time-based:
   returns exactly `NumValuesPerNode` values, and the server stores a resume cursor and returns an opaque continuation
   point (16 bytes). The client hands it back to fetch the next page.
 - When it does not, the page is the **last** — no continuation point, even when it is exactly full. (#692: a total that
-  is an exact multiple of the cap used to end with an extra empty `GoodNoData` page.)
+  is an exact multiple of the cap used to end with an extra empty `GoodNoData` page.) The exception is an oversized tie
+  cluster (below): its pages are sliced from the over-fetched cluster, its last slice is `min(N, remaining)` values and
+  still carries a continuation point when the window goes on — so stepping past a cluster at the very end of the data can
+  yield one empty final page.
 - `NumValuesPerNode == 0` ("all values, no limit") is never paged; the whole window returns in one
   shot.
 - The resume cursor is **tie-safe**: the next page resumes from the last returned sample's
   SourceTimestamp *inclusive* and drops the boundary samples already emitted, so samples sharing
   the boundary timestamp are neither duplicated nor skipped. Because it re-reads them, a resumed page asks for
-  `NumValuesPerNode + skip + 1` rows, so it still carries a full page.
-- ⚠️ **#692 (client-visible): every page is full.** A resumed page used to ask for only `NumValuesPerNode` rows, so after
+  `NumValuesPerNode + skip + 1` rows, so it still carries a full page. On a strict Gateway
+  (`AllowStreamingBeyondCeiling=false`) that moves the row-ceiling edge by `skip + 1`: a `NumValuesPerNode` just under the
+  learned ceiling may now cost one learning refusal (#691.3 re-asks at the ceiling). A page clamp to the learned ceiling is
+  being added separately (#694).
+- ⚠️ **#692 (client-visible): every page is full** (outside an oversized tie cluster — see above). A resumed page used to ask for only `NumValuesPerNode` rows, so after
   the trim it netted `N − skip` values — and at `NumValuesPerNode = 1` **nothing**: the lone re-read boundary sample was
   trimmed, the empty "full" page was taken for a stalled tie cluster, and a `MaxTieClusterOverfetch + 1` (65,537-row)
   over-fetch stepped one tick to return an **empty page with a continuation point between every value** (live, AVEVA:
@@ -610,7 +616,10 @@ paging time-based:
 > `ServerHistorian:MaxTieClusterOverfetch` (default **65 536**), then serves the cluster
 > `NumValuesPerNode` samples at a time across successive pages, advancing the cursor one tick past
 > the timestamp once the cluster is fully drained. A cursor left inside the cluster is marked so (`WithinTieCluster`), and
-> its next page over-fetches the cluster directly instead of first re-reading rows it knows are ties.
+> its next page over-fetches the cluster directly instead of first re-reading rows it knows are ties. A drained cluster
+> resumes only while the next step is strictly before `EndTime` (truncated to the backend's resolution) — the forward
+> window is end-exclusive; resuming AT `EndTime` used to make the next read a point read that returned the samples at
+> `EndTime` as data (#692 review).
 >
 > **Reverse reads** (`start > end`, newest first) page the same way in the opposite direction: a drained
 > cluster steps the cursor one tick **down**, the read ends when that would reach the (excluded) `end`, and
diff --git a/src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/HistoryPaging.cs b/src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/HistoryPaging.cs
index 28082027..c16ba97b 100644
--- a/src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/HistoryPaging.cs
+++ b/src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/HistoryPaging.cs
@@ -258,12 +258,11 @@ internal static class HistoryPaging
     /// <param name="cap">The client's per-page cap (<c>NumValuesPerNode</c>); must be &gt; 0 to make progress.</param>
     /// <param name="boundaryT">The single SourceTimestamp every tie in the cluster shares.</param>
     /// <param name="endUtc">The (inclusive) upper bound of the read window; unchanged across pages.
-    ///     OpcUaServer-003, accepted by design: the backend contract
-    ///     (<c>IHistorianDataSource.ReadRawAsync</c>) treats <c>endUtc</c> as EXCLUSIVE, so the
-    ///     <c>next &lt;= endUtc</c> test here is deliberately the conservative side — at worst one extra
-    ///     empty resume page (<c>[endUtc, endUtc)</c> → GoodNoData, no continuation point), never a
-    ///     dropped or duplicated sample. Tightening to <c>&lt;</c> would be correct only if the live
-    ///     backend's boundary is confirmed exclusive; it is not worth the opposite off-by-one risk.</param>
+    ///     The forward window is end-EXCLUSIVE (Part 11; the <c>IHistorianDataSource.ReadRawAsync</c> contract), so a drained
+    ///     cluster continues only while <c>next</c> is strictly before <paramref name="endUtc"/> truncated to
+    ///     <paramref name="resolutionTicks"/>. (OpcUaServer-003 had kept <c>next &lt;= endUtc</c> as "the conservative side",
+    ///     but it was not: resuming AT EndTime makes the next read a <c>start == end</c> point read, which returns the samples
+    ///     at EndTime as data — #692 review F1; on a microsecond backend EndTime is truncated first.)</param>
     /// <param name="sliceStart">The index into the cluster the emitted slice starts at (clamped to the count).</param>
     /// <param name="sliceCount">How many ties the emitted slice contains (may be <c>0</c> on a stale-skip self-heal).</param>
     /// <param name="nextStartUtc">The next page's inclusive start: <paramref name="boundaryT"/> while the
@@ -306,7 +305,9 @@ internal static class HistoryPaging
         else
         {
             var next = boundaryT.AddTicks(resolutionTicks);
-            if (next <= endUtc) { nextStartUtc = next; nextSkip = 0; }
+            // The forward window is end-EXCLUSIVE, at the backend's resolution: resuming AT EndTime would turn the next read
+            // into a start == end point read, which returns the samples at EndTime as data (#692 review F1).
+            if (next.Ticks < endUtc.Ticks - endUtc.Ticks % resolutionTicks) { nextStartUtc = next; nextSkip = 0; }
             else { nextStartUtc = null; nextSkip = 0; }
         }
     }
diff --git a/tests/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests/NodeManagerHistoryReadPageFillTests.cs b/tests/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests/NodeManagerHistoryReadPageFillTests.cs
index 36c7cf3e..87ae0cde 100644
--- a/tests/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests/NodeManagerHistoryReadPageFillTests.cs
+++ b/tests/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer.Tests/NodeManagerHistoryReadPageFillTests.cs
@@ -146,6 +146,68 @@ public sealed class NodeManagerHistoryReadPageFillTests(BoundsServerFixture fx)
         pages.Select(p => p.Count).ShouldBe([2, 2, 2, 2]);
     }
 
+    /// <summary>
+    ///     #692 review F1: an oversized tie cluster one step below EndTime used to resume AT EndTime (the drain check was
+    ///     <c>next &lt;= end</c>, but the forward window is end-exclusive), so the next read was a <c>start == end</c> point
+    ///     read that returned the samples AT EndTime as data. Paged must equal unpaged — neither holds the EndTime samples.
+    ///     The microsecond shape truncates EndTime (8.1 µs ⇒ 8.0 µs) before the same thing happens.
+    /// </summary>
+    [Theory]
+    [InlineData(Shape.Native, 1L, 8L)]
+    [InlineData(Shape.Microsecond, 10L, 81L)]
+    public async Task A_tie_cluster_just_below_the_end_time_never_resumes_onto_the_end_time(Shape shape, long unit, long endTicks)
+    {
+        var series = new List<DataValueSnapshot>();
+        var v = 0;
+        foreach (var (at, count) in new[] { (5L, 5), (6L, 2), (7L, 3), (8L, 3) })
+            for (var i = 0; i < count; i++)
+                series.Add(Sample(v++, T0.AddTicks(at * unit)));
+        var source = Source(shape, series);
+        var (start, end) = (T0.AddSeconds(-3), T0.AddTicks(endTicks));
+
+        var unpaged = await SinglePageAsync(source, start, end, bounds: false);
+        unpaged.Count.ShouldBe(10, "the samples at the (exclusive) end time are not data");
+
+        var pages = await DrainPagesAsync(source, start, end, 1, bounds: false);
+        Render(pages.SelectMany(p => p)).ShouldBe(Render(unpaged));
+    }
+
+    /// <summary>
+    ///     #692: a cursor left inside an oversized tie cluster is flagged, and its next page goes straight to the cluster
+    ///     over-fetch — one backend read — instead of first re-reading <c>N + skip + 1</c> rows it knows are ties.
+    /// </summary>
+    [Fact]
+    public async Task A_cursor_inside_a_tie_cluster_over_fetches_without_a_normal_read_first()
+    {
+        // a @1s, five ties @2s, b @3s; N = 1.
+        var series = new List<DataValueSnapshot> { Sample(0, T0.AddSeconds(1)) };
+        for (var i = 1; i <= 5; i++) series.Add(Sample(i, T0.AddSeconds(2)));
+        series.Add(Sample(6, T0.AddSeconds(3)));
+        var source = Source(Shape.Native, series);
+        var (start, end) = (T0, T0.AddSeconds(10));
+
+        var readsPerPage = new List<int>();
+        var all = new List<DataValue>();
+        byte[]? cp = null;
+        do
+        {
+            source.ResetCounts();
+            var (values, error, next) = await ReadAsync(source, start, end, 1, bounds: false, cp);
+            error.StatusCode.Code.ShouldBe(StatusCodes.Good);
+            values.Count.ShouldBe(1);
+            all.AddRange(values);
+            readsPerPage.Add(source.Reads);
+            cp = next;
+            readsPerPage.Count.ShouldBeLessThan(20);
+        }
+        while (cp is not null);
+
+        all.Select(x => (double)x.Value).ShouldBe([0d, 1d, 2d, 3d, 4d, 5d, 6d]);
+        // Page 1 [a], page 2 [t1]: one read each. Page 3 [t2]: the read stalls on the cluster, then the over-fetch (2).
+        // Pages 4 [t3], 5 [t4], 6 [t5]: the flagged cursor over-fetches directly (1 each). Page 7 [b]: one read.
+        readsPerPage.ShouldBe([1, 1, 2, 1, 1, 1, 1]);
+    }
+
     /// <summary>
     ///     #692 (low): a one-value bounded read whose start bound is not found asks whether the read holds anything. When the
     ///     answer is "only the end bound", the bound it just found rides the continuation point — the next page is served