diff --git a/cpp/matching_engine_cpp.cpp b/cpp/matching_engine_cpp.cpp
index 245d267..008f97c 100644
--- a/cpp/matching_engine_cpp.cpp
+++ b/cpp/matching_engine_cpp.cpp
@@ -7,6 +7,16 @@ namespace py = pybind11;
 
 namespace {
 
+struct BatchBuffers {
+    py::ssize_t n;
+    const int64_t* ts_ns;
+    const uint8_t* action;
+    const uint8_t* side;
+    const int64_t* price_ticks;
+    const int32_t* size;
+    const int64_t* order_id;
+};
+
 template <typename T>
 const T* checked_buffer(
     const py::buffer& buffer,
@@ -31,8 +41,7 @@ const T* checked_buffer(
     return static_cast<const T*>(info.ptr);
 }
 
-std::vector<FillRow> apply_events_batch(
-    MatchingEngineCpp& engine,
+BatchBuffers read_batch_buffers(
     const py::buffer& ts_ns,
     const py::buffer& action,
     const py::buffer& side,
@@ -53,48 +62,70 @@ std::vector<FillRow> apply_events_batch(
     }
 
     const py::ssize_t n = ts_info.shape[0];
-    const auto* ts_data = static_cast<const int64_t*>(ts_info.ptr);
-    const auto* action_data = checked_buffer<uint8_t>(
+    return BatchBuffers{
+        n,
+        static_cast<const int64_t*>(ts_info.ptr),
+        checked_buffer<uint8_t>(
+            action,
+            py::format_descriptor<uint8_t>::format().c_str(),
+            "action",
+            n
+        ),
+        checked_buffer<uint8_t>(
+            side,
+            py::format_descriptor<uint8_t>::format().c_str(),
+            "side",
+            n
+        ),
+        checked_buffer<int64_t>(
+            price_ticks,
+            py::format_descriptor<int64_t>::format().c_str(),
+            "price_ticks",
+            n
+        ),
+        checked_buffer<int32_t>(
+            size,
+            py::format_descriptor<int32_t>::format().c_str(),
+            "size",
+            n
+        ),
+        checked_buffer<int64_t>(
+            order_id,
+            py::format_descriptor<int64_t>::format().c_str(),
+            "order_id",
+            n
+        ),
+    };
+}
+
+std::vector<FillRow> apply_events_batch(
+    MatchingEngineCpp& engine,
+    const py::buffer& ts_ns,
+    const py::buffer& action,
+    const py::buffer& side,
+    const py::buffer& price_ticks,
+    const py::buffer& size,
+    const py::buffer& order_id
+) {
+    const BatchBuffers batch = read_batch_buffers(
+        ts_ns,
         action,
-        py::format_descriptor<uint8_t>::format().c_str(),
-        "action",
-        n
-    );
-    const auto* side_data = checked_buffer<uint8_t>(
         side,
-        py::format_descriptor<uint8_t>::format().c_str(),
-        "side",
-        n
-    );
-    const auto* price_data = checked_buffer<int64_t>(
         price_ticks,
-        py::format_descriptor<int64_t>::format().c_str(),
-        "price_ticks",
-        n
-    );
-    const auto* size_data = checked_buffer<int32_t>(
         size,
-        py::format_descriptor<int32_t>::format().c_str(),
-        "size",
-        n
-    );
-    const auto* order_data = checked_buffer<int64_t>(
-        order_id,
-        py::format_descriptor<int64_t>::format().c_str(),
-        "order_id",
-        n
+        order_id
     );
 
     std::vector<FillRow> fills;
 
-    for (py::ssize_t i = 0; i < n; ++i) {
+    for (py::ssize_t i = 0; i < batch.n; ++i) {
         const auto event_fills = engine.apply_event_code(
-            ts_data[i],
-            static_cast<char>(action_data[i]),
-            static_cast<char>(side_data[i]),
-            price_data[i],
-            size_data[i],
-            order_data[i]
+            batch.ts_ns[i],
+            static_cast<char>(batch.action[i]),
+            static_cast<char>(batch.side[i]),
+            batch.price_ticks[i],
+            batch.size[i],
+            batch.order_id[i]
         );
         fills.insert(fills.end(), event_fills.begin(), event_fills.end());
     }
@@ -102,6 +133,44 @@ std::vector<FillRow> apply_events_batch(
     return fills;
 }
 
+py::tuple apply_events_until_fill(
+    MatchingEngineCpp& engine,
+    const py::buffer& ts_ns,
+    const py::buffer& action,
+    const py::buffer& side,
+    const py::buffer& price_ticks,
+    const py::buffer& size,
+    const py::buffer& order_id
+) {
+    const BatchBuffers batch = read_batch_buffers(
+        ts_ns,
+        action,
+        side,
+        price_ticks,
+        size,
+        order_id
+    );
+
+    std::vector<FillRow> fills;
+
+    for (py::ssize_t i = 0; i < batch.n; ++i) {
+        const auto event_fills = engine.apply_event_code(
+            batch.ts_ns[i],
+            static_cast<char>(batch.action[i]),
+            static_cast<char>(batch.side[i]),
+            batch.price_ticks[i],
+            batch.size[i],
+            batch.order_id[i]
+        );
+        if (!event_fills.empty()) {
+            fills.insert(fills.end(), event_fills.begin(), event_fills.end());
+            return py::make_tuple(i + 1, fills);
+        }
+    }
+
+    return py::make_tuple(batch.n, fills);
+}
+
 }  // namespace
 
 PYBIND11_MODULE(_matching_engine_cpp, module) {
@@ -116,6 +185,7 @@ PYBIND11_MODULE(_matching_engine_cpp, module) {
         .def(py::init<>())
         .def("apply_event", &MatchingEngineCpp::apply_event)
         .def("apply_events_batch", &apply_events_batch)
+        .def("apply_events_until_fill", &apply_events_until_fill)
         .def("place_limit", &MatchingEngineCpp::place_limit)
         .def("place_market", &MatchingEngineCpp::place_market)
         .def("cancel", &MatchingEngineCpp::cancel)
diff --git a/docs/benchmarks.md b/docs/benchmarks.md
index 7ccbcab..73512c8 100644
--- a/docs/benchmarks.md
+++ b/docs/benchmarks.md
@@ -88,5 +88,10 @@ up the auditability of `ReplayResult`. It also keeps fill-connected strategies
 honest: Python should regain control when execution state changes in a way the
 strategy can observe.
 
+The first engine primitive for that design is
+`CppMatchingEngine.apply_events_until_fill(...)`. It consumes a compiled slice
+until the first passive fill boundary and returns the event count needed to
+resume replay at the correct row.
+
 CI should keep benchmark code runnable; it should not enforce fixed speed
 thresholds across hardware.
diff --git a/docs/execution-engines.md b/docs/execution-engines.md
index 3e9a9ab..b8e59d6 100644
--- a/docs/execution-engines.md
+++ b/docs/execution-engines.md
@@ -99,6 +99,23 @@ engine = CppMatchingEngine(tick_size=spec.tick_size)
 fills = engine.apply_events_batch(columns.slice(0, len(events)))
 ```
 
+The wrapper also exposes `apply_events_until_fill(...)`, which is the first
+low-level primitive for boundary-batched replay. It lets the C++ engine consume
+a compiled market-data slice independently until either the slice ends or a
+passive fill occurs. The method returns both the number of events consumed and
+the fills produced at the boundary, so a future replay loop can resume Python at
+the exact point where strategy-visible execution state changed.
+
+```python
+events_consumed, fills = engine.apply_events_until_fill(
+    columns.slice(start, stop),
+)
+```
+
+This primitive does not change ordinary audited `Replay(...)` yet. It exists so
+the faster path can be wired in carefully without weakening replay
+inspectability.
+
 Install a source checkout normally to build the extension:
 
 ```bash
diff --git a/src/ordersim/sim/cpp_matching_engine.py b/src/ordersim/sim/cpp_matching_engine.py
index 18dae8d..cbbf390 100644
--- a/src/ordersim/sim/cpp_matching_engine.py
+++ b/src/ordersim/sim/cpp_matching_engine.py
@@ -53,6 +53,22 @@ def apply_events_batch(
         )
         return [self._fill_from_row(row) for row in rows]
 
+    def apply_events_until_fill(
+        self,
+        events: CompiledEventSlice,
+    ) -> tuple[int, list[Fill]]:
+        """Apply a compiled slice until the first passive fill boundary."""
+
+        events_applied, rows = self._core.apply_events_until_fill(
+            events.ts_ns,
+            events.action,
+            events.side,
+            events.price_ticks,
+            events.size,
+            events.order_id,
+        )
+        return int(events_applied), [self._fill_from_row(row) for row in rows]
+
     def advance_time(self, ts_ns: int) -> None:
         self._core.advance_time(ts_ns)
 
diff --git a/tests/test_cpp_execution_engine.py b/tests/test_cpp_execution_engine.py
index 164b8a6..36b90ec 100644
--- a/tests/test_cpp_execution_engine.py
+++ b/tests/test_cpp_execution_engine.py
@@ -131,6 +131,87 @@ def test_cpp_execution_engine_applies_compiled_event_batches() -> None:
     ]
 
 
+def test_cpp_execution_engine_stops_compiled_batch_at_passive_fill() -> None:
+    events = (
+        MBOEvent(
+            ts_ns=1,
+            action="add",
+            side="bid",
+            price=Decimal("100.0"),
+            size=1,
+            order_id=1,
+        ),
+        MBOEvent(
+            ts_ns=2,
+            action="trade",
+            side="bid",
+            price=Decimal("100.0"),
+            size=2,
+            order_id=2,
+        ),
+        MBOEvent(
+            ts_ns=3,
+            action="add",
+            side="ask",
+            price=Decimal("101.0"),
+            size=1,
+            order_id=3,
+        ),
+    )
+    columns = CompiledEventColumns.from_events(events, tick_size=Decimal("0.10"))
+    engine = CppMatchingEngine(tick_size=Decimal("0.10"))
+
+    resting = engine.place_limit(side="buy", price=Decimal("100.0"), size=1)
+    events_applied, fills = engine.apply_events_until_fill(
+        columns.slice(0, len(events))
+    )
+
+    assert resting.order_id is not None
+    assert events_applied == 2
+    assert [(fill.order_id, fill.ts_ns, fill.size) for fill in fills] == [
+        (resting.order_id, 2, 1),
+    ]
+    assert engine.book_top() == (None, None)
+
+    remaining_fills = engine.apply_events_batch(
+        columns.slice(events_applied, len(events))
+    )
+
+    assert remaining_fills == []
+    assert engine.book_top() == (None, Decimal("101.00"))
+
+
+def test_cpp_execution_engine_reports_full_batch_without_passive_fill() -> None:
+    events = (
+        MBOEvent(
+            ts_ns=1,
+            action="add",
+            side="bid",
+            price=Decimal("100.0"),
+            size=1,
+            order_id=1,
+        ),
+        MBOEvent(
+            ts_ns=2,
+            action="add",
+            side="ask",
+            price=Decimal("101.0"),
+            size=1,
+            order_id=2,
+        ),
+    )
+    columns = CompiledEventColumns.from_events(events, tick_size=Decimal("0.10"))
+    engine = CppMatchingEngine(tick_size=Decimal("0.10"))
+
+    events_applied, fills = engine.apply_events_until_fill(
+        columns.slice(0, len(events))
+    )
+
+    assert events_applied == len(events)
+    assert fills == []
+    assert engine.book_top() == (Decimal("100.00"), Decimal("101.00"))
+
+
 def test_compiled_event_columns_reject_unaligned_prices() -> None:
     events = (
         MBOEvent(