diff --git a/app/modules/services/default_task_sync.py b/app/modules/services/default_task_sync.py
index 75bb93a..2b19d40 100644
--- a/app/modules/services/default_task_sync.py
+++ b/app/modules/services/default_task_sync.py
@@ -176,6 +176,36 @@ def sync_firm_tasks_from_system_defaults(
         firm_task_ids=[int(row.id) for row in firm_rows if getattr(row, "id", None)],
     )
 
+    # ------------------------------------------------------------------
+    # IMPORTANT: two-phase sequence re-numbering
+    #
+    # The database enforces a UNIQUE constraint on:
+    #   (tenant_id, service_catalogue_id, sequence_no)
+    #
+    # Existing duplicate imports may contain several generations of the same
+    # task at different sequence numbers. During a sync, moving a canonical
+    # row directly onto the latest system-default sequence can therefore
+    # collide with another historical row that still owns that sequence.
+    #
+    # Marking that historical row inactive does NOT release the unique key.
+    # Park every existing firm template on a guaranteed-unique temporary
+    # sequence first, flush, and only then apply final/default sequences.
+    # ------------------------------------------------------------------
+    max_existing_sequence = max(
+        [int(getattr(row, "sequence_no", 0) or 0) for row in firm_rows] + [0]
+    )
+    max_default_sequence = max(
+        [int(getattr(row, "sequence_no", 0) or 0) for row in defaults] + [0]
+    )
+    parking_base = max(max_existing_sequence, max_default_sequence, 0) + 100000
+
+    for offset, row in enumerate(sorted(firm_rows, key=lambda item: int(item.id)), start=1):
+        row.sequence_no = parking_base + offset
+
+    # This flush is intentional and must happen before any canonical row is
+    # assigned a system-default sequence number.
+    db.flush()
+
     # First clean pre-existing duplicate groups, regardless of whether the task
     # remains in the current system defaults. A unique firm-only task is untouched.
     for candidates in by_name.values():
@@ -188,6 +218,10 @@ def sync_firm_tasks_from_system_defaults(
         for duplicate in candidates:
             if duplicate.id == canonical.id:
                 continue
+
+            # Duplicate rows are retained for historical FK safety but remain
+            # on their unique parked sequence number. The UNIQUE sequence
+            # constraint therefore remains satisfied even after deactivation.
             if getattr(duplicate, "is_active", True):
                 duplicate.is_active = False
                 if (
@@ -221,28 +255,13 @@ def sync_firm_tasks_from_system_defaults(
 
         # No name match. A unique sequence match is a conservative fallback for a
         # system-default rename while still avoiding arbitrary replacement.
-        sequence_no = getattr(default, "sequence_no", None)
-        sequence_candidates = [
-            row
-            for row in firm_rows
-            if getattr(row, "sequence_no", None) == sequence_no
-            and getattr(row, "is_active", True)
-        ]
-        if len(sequence_candidates) == 1:
-            target = sequence_candidates[0]
-            changed = _copy_default_columns(default, target)
-            if (
-                updated_by_user_id is not None
-                and hasattr(target, "updated_by_user_id")
-            ):
-                target.updated_by_user_id = updated_by_user_id
-            if changed:
-                result.updated += 1
-            else:
-                result.unchanged += 1
-            # Make subsequent defaults see the new name.
-            by_name[_normalise_name(getattr(target, "task_name", None))].append(target)
-            continue
+        # Do NOT use the pre-sync sequence number as an identity fallback here.
+        # All existing rows have deliberately been parked on temporary sequence
+        # numbers to satisfy the DB unique constraint. More importantly,
+        # sequence number is ordering metadata, not a stable task identity.
+        # A renamed system-default task without a name match is therefore
+        # treated as a new task rather than risking replacement of the wrong
+        # firm-specific task.
 
         # Missing task: construct from the intersection of mapped columns.
         source_cols = _column_names(ServiceDefaultTaskTemplate)