diff --git a/erpnext/manufacturing/doctype/work_order/test_work_order.py b/erpnext/manufacturing/doctype/work_order/test_work_order.py
index b4f504f1adc..c5f92178d7f 100644
--- a/erpnext/manufacturing/doctype/work_order/test_work_order.py
+++ b/erpnext/manufacturing/doctype/work_order/test_work_order.py
@@ -3414,6 +3414,68 @@ class TestWorkOrder(ERPNextTestSuite):
 		self.assertEqual(flt(disassembly_row.transfer_qty), expected_disassembly_qty)
 		disassembly.submit()
 
+	def test_disassembly_representative_is_stable_when_entries_share_a_creation(self):
+		"""The representative line must be decided by the query, not by row order on disk.
+
+		When two Manufacture entries share a creation timestamp the entry name decides, as it always
+		did -- but compared in Python, so the database's collation does not.
+		"""
+		from erpnext.stock.doctype.stock_entry.services.disassemble import DisassembleStockEntry
+		from erpnext.stock.doctype.stock_entry.test_stock_entry import (
+			make_stock_entry as make_stock_entry_test_record,
+		)
+
+		raw_item = make_item("Test RM for Disassembly Tie", {"is_stock_item": 1}).name
+		fg_item = make_item("Test FG for Disassembly Tie", {"is_stock_item": 1}).name
+		bom = make_bom(item=fg_item, quantity=1, raw_materials=[raw_item], rm_qty=2)
+
+		wo = make_wo_order_test_record(production_item=fg_item, qty=10, bom_no=bom.name, status="Not Started")
+		make_stock_entry_test_record(
+			item_code=raw_item,
+			purpose="Material Receipt",
+			target=wo.wip_warehouse,
+			qty=50,
+			basic_rate=100,
+		)
+		transfer = frappe.get_doc(make_stock_entry(wo.name, "Material Transfer for Manufacture", wo.qty))
+		for item in transfer.items:
+			item.s_warehouse = wo.wip_warehouse
+		transfer.save()
+		transfer.submit()
+
+		first = frappe.get_doc(make_stock_entry(wo.name, "Manufacture", 5))
+		first.submit()
+		second = frappe.get_doc(make_stock_entry(wo.name, "Manufacture", 5))
+		second.submit()
+
+		shared_creation = frappe.db.get_value("Stock Entry", first.name, "creation")
+		frappe.db.set_value("Stock Entry", second.name, "creation", shared_creation, update_modified=False)
+
+		rows = frappe.get_all(
+			"Stock Entry Detail",
+			filters={"parent": ("in", [first.name, second.name]), "item_code": raw_item},
+			fields=["name", "parent", "idx"],
+			order_by="name",
+		)
+		self.assertEqual(len(rows), 2)
+		self.assertEqual(len({row.idx for row in rows}), 1, "idx must repeat for the tie to matter")
+
+		# mark the two lines apart on a column the representative alone supplies
+		warehouses = {}
+		for offset, row in enumerate(rows):
+			warehouse = create_warehouse(f"_Test Disassembly Tie {offset}")
+			warehouses[row.parent] = warehouse
+			frappe.db.set_value(
+				"Stock Entry Detail", row.name, "s_warehouse", warehouse, update_modified=False
+			)
+
+		service = DisassembleStockEntry(frappe._dict(work_order=wo.name, source_stock_entry=None))
+		source_row = next(
+			row for row in service.get_items_from_manufacture_stock_entry() if row.item_code == raw_item
+		)
+
+		self.assertEqual(source_row.s_warehouse, warehouses[min(warehouses, key=str.casefold)])
+
 	def test_disassembly_with_additional_rm_not_in_bom(self):
 		"""
 		Test that SE-linked disassembly includes additional raw materials
diff --git a/erpnext/stock/doctype/stock_entry/services/disassemble.py b/erpnext/stock/doctype/stock_entry/services/disassemble.py
index 0b31822141d..fef1b08a107 100644
--- a/erpnext/stock/doctype/stock_entry/services/disassemble.py
+++ b/erpnext/stock/doctype/stock_entry/services/disassemble.py
@@ -24,6 +24,10 @@ def _qty_tolerance(precision: int) -> float:
 	return 1.0 / (10**precision)
 
 
+def _manufacture_line_order(line):
+	return (line.parent_creation, line.parent_name.casefold(), line.parent_name, line.idx)
+
+
 class DisassembleStockEntry(BaseStockEntry):
 	def validate(self):
 		self.validate_warehouse()
@@ -382,6 +386,10 @@ class DisassembleStockEntry(BaseStockEntry):
 		is_finished_item decides whether the row is the output or an input. Aggregating each column
 		on its own can pair values from different lines into a row that was never posted, so take
 		the columns from a single real line instead. UOM is normalized separately to stock UOM.
+
+		Ordered in Python rather than SQL, keeping the original (entry creation, entry name, line idx)
+		precedence. casefold reproduces MariaDB's case-insensitive collation without depending on the
+		database's, which PostgreSQL resolves by byte value.
 		"""
 		SE = frappe.qb.DocType("Stock Entry")
 		SED = frappe.qb.DocType("Stock Entry Detail")
@@ -391,6 +399,9 @@ class DisassembleStockEntry(BaseStockEntry):
 			.join(SE)
 			.on(SED.parent == SE.name)
 			.select(
+				SE.creation.as_("parent_creation"),
+				SE.name.as_("parent_name"),
+				SED.idx,
 				SED.item_code,
 				SED.item_name,
 				SED.description,
@@ -409,14 +420,14 @@ class DisassembleStockEntry(BaseStockEntry):
 			.where(
 				(SE.docstatus == 1) & (SE.purpose == "Manufacture") & (SE.work_order == self.doc.work_order)
 			)
-			.orderby(SE.creation)
-			.orderby(SE.name)
-			.orderby(SED.idx)
 			.run(as_dict=True)
 		)
+		lines.sort(key=_manufacture_line_order)
 
 		representative = {}
 		for line in lines:
+			line.pop("parent_creation")
+			line.pop("parent_name")
 			representative.setdefault(line.item_code, line)
 
 		return representative