diff --git a/src/pyrecest/backend_support/_shared_numpy_assignment_by_sum_contract.py b/src/pyrecest/backend_support/_shared_numpy_assignment_by_sum_contract.py
new file mode 100644
index 000000000..6a6f89925
--- /dev/null
+++ b/src/pyrecest/backend_support/_shared_numpy_assignment_by_sum_contract.py
@@ -0,0 +1,81 @@
+"""Shared NumPy ``assignment_by_sum`` duplicate-index compatibility hook."""
+
+from __future__ import annotations
+
+import sys
+
+
+def _install_assignment_by_sum(assignment_by_sum, backend, shared_numpy) -> None:
+    """Install the patched helper on loaded shared-NumPy facade modules."""
+
+    shared_numpy.assignment_by_sum = assignment_by_sum
+    for module_name in ("pyrecest._backend.numpy", "pyrecest._backend.autograd"):
+        module = sys.modules.get(module_name)
+        if module is not None:
+            module.assignment_by_sum = assignment_by_sum
+    if getattr(backend, "__backend_name__", None) in {"numpy", "autograd"}:
+        backend.assignment_by_sum = assignment_by_sum
+
+
+def patch_shared_numpy_assignment_by_sum_duplicate_indices() -> None:
+    """Make shared NumPy assignment-by-sum accumulate repeated indices."""
+
+    try:
+        import pyrecest._backend._shared_numpy as shared_numpy  # pylint: disable=import-outside-toplevel
+        import pyrecest.backend as backend  # pylint: disable=import-outside-toplevel
+    except ModuleNotFoundError:  # pragma: no cover - backend import failure path
+        return
+
+    if getattr(backend, "__backend_name__", None) not in {"numpy", "autograd"}:
+        return
+
+    original_assignment_by_sum = shared_numpy.assignment_by_sum
+    if getattr(
+        original_assignment_by_sum,
+        "_pyrecest_duplicate_index_accumulation_contract",
+        False,
+    ):
+        _install_assignment_by_sum(original_assignment_by_sum, backend, shared_numpy)
+        return
+
+    def assignment_by_sum(x, values, indices, axis=0):
+        x_new = shared_numpy.copy(shared_numpy.array(x))
+
+        if shared_numpy._is_empty_index_sequence(indices):
+            return x_new
+
+        use_vectorization = hasattr(indices, "__len__") and len(indices) < shared_numpy.ndim(
+            x_new
+        )
+        if shared_numpy._is_boolean(indices):
+            x_new[indices] += values
+            return x_new
+
+        zip_indices = shared_numpy._is_iterable(indices) and shared_numpy._is_iterable(
+            indices[0]
+        )
+        if zip_indices:
+            indices = tuple(zip(*indices))
+        if not use_vectorization:
+            len_indices = len(indices) if shared_numpy._is_iterable(indices) else 1
+            len_values = len(values) if shared_numpy._is_iterable(values) else 1
+            if len_values > 1 and len_values != len_indices:
+                raise ValueError("Either one value or as many values as indices")
+            shared_numpy._np.add.at(x_new, indices, values)
+        else:
+            indices = tuple(list(indices[:axis]) + [slice(None)] + list(indices[axis:]))
+            x_new[indices] += values
+        return x_new
+
+    assignment_by_sum.__name__ = getattr(
+        original_assignment_by_sum,
+        "__name__",
+        "assignment_by_sum",
+    )
+    assignment_by_sum.__doc__ = getattr(original_assignment_by_sum, "__doc__", None)
+    assignment_by_sum._pyrecest_duplicate_index_accumulation_contract = True
+
+    _install_assignment_by_sum(assignment_by_sum, backend, shared_numpy)
+
+
+__all__ = ["patch_shared_numpy_assignment_by_sum_duplicate_indices"]
diff --git a/src/pyrecest/stability.py b/src/pyrecest/stability.py
index aa56e78f3..9e6e8cb0f 100644
--- a/src/pyrecest/stability.py
+++ b/src/pyrecest/stability.py
@@ -9,8 +9,12 @@
 from pyrecest.backend_support._pytorch_allclose_device_contract import (
     patch_pytorch_allclose_device_contract as _patch_pytorch_allclose_device_contract,
 )
+from pyrecest.backend_support._shared_numpy_assignment_by_sum_contract import (
+    patch_shared_numpy_assignment_by_sum_duplicate_indices as _patch_shared_numpy_assignment_by_sum_duplicate_indices,
+)
 
 _patch_pytorch_allclose_device_contract()
+_patch_shared_numpy_assignment_by_sum_duplicate_indices()
 
 P = ParamSpec("P")
 R = TypeVar("R")
diff --git a/tests/backend_support/test_shared_numpy_assignment_by_sum_duplicate_indices.py b/tests/backend_support/test_shared_numpy_assignment_by_sum_duplicate_indices.py
new file mode 100644
index 000000000..79c73fa2c
--- /dev/null
+++ b/tests/backend_support/test_shared_numpy_assignment_by_sum_duplicate_indices.py
@@ -0,0 +1,31 @@
+import importlib.util
+
+import pytest
+
+from tests.support.backend_runner import run_backend_code
+
+pytestmark = pytest.mark.backend_portable
+
+
+@pytest.mark.parametrize("backend_name", ["numpy", "autograd"])
+def test_assignment_by_sum_accumulates_duplicate_advanced_indices(backend_name):
+    if backend_name == "autograd" and importlib.util.find_spec("autograd") is None:
+        pytest.skip("autograd is not installed")
+
+    code = """
+import pyrecest.backend as backend
+
+vector = backend.zeros(3)
+vector_result = backend.assignment_by_sum(vector, [1.0, 2.0], [0, 0])
+assert backend.to_numpy(vector_result).tolist() == [3.0, 0.0, 0.0]
+
+matrix = backend.zeros((2, 2))
+matrix_result = backend.assignment_by_sum(matrix, [1.0, 2.0], [(0, 1), (0, 1)])
+assert backend.to_numpy(matrix_result).tolist() == [[0.0, 3.0], [0.0, 0.0]]
+
+print("ok")
+"""
+    result = run_backend_code(backend_name, code)
+
+    assert result.returncode == 0, result.stderr
+    assert "ok" in result.stdout