diff --git a/backend/app/simulation.py b/backend/app/simulation.py
index 74be6e0..97d0da4 100644
--- a/backend/app/simulation.py
+++ b/backend/app/simulation.py
@@ -133,10 +133,16 @@ def allocate_capital(
 
     used = set()
 
-    def _fill(bucket_idx: int, eligible: list[Candidate], require_dividend: bool):
+    def _fill(bucket_idx: int, eligible: list[Candidate], require_dividend: bool,
+              sort_by: str = "combined_score"):
         nonlocal cash, used
         remaining = cash[bucket_idx]
-        for cand in sorted(eligible, key=lambda c: -c.combined_score):
+        # bucket 3 must rank by dividend_yield (ignoring score); others by score.
+        if sort_by == "dividend_yield":
+            key = lambda c: -c.dividend_yield
+        else:
+            key = lambda c: -c.combined_score
+        for cand in sorted(eligible, key=key):
             if cand.symbol in used:
                 continue
             if require_dividend and not cand.is_dividend:
@@ -164,8 +170,8 @@ def allocate_capital(
     _fill(0, [c for c in by_score if c.is_dividend], require_dividend=True)
     # Bucket 2: non-dividend, highest score
     _fill(1, [c for c in by_score if not c.is_dividend], require_dividend=False)
-    # Bucket 3: highest dividend yield, excluding symbols already bought
-    _fill(2, by_yield, require_dividend=True)
+    # Bucket 3: highest dividend yield (ignoring score), excluding symbols bought
+    _fill(2, by_yield, require_dividend=True, sort_by="dividend_yield")
 
     result.unallocated_cash = sum(cash)
     return result
diff --git a/backend/tests/test_simulation.py b/backend/tests/test_simulation.py
index 8614dd7..71ca943 100644
--- a/backend/tests/test_simulation.py
+++ b/backend/tests/test_simulation.py
@@ -50,6 +50,25 @@ class AllocationTest(unittest.TestCase):
         overlap = set(bucket3_syms) & set(bucket1_syms)
         self.assertEqual(overlap, set())
 
+    def test_bucket3_ranks_by_yield_ignoring_score(self):
+        # The spec: bucket3 = highest dividend yield, IGNORING score.
+        # A low-score but high-yield name must rank above a high-score low-yield name.
+        cands = [
+            Candidate("HIGH_SCORE", 10.0, 8.0, True, 2.0),   # score 8, yield 2%
+            Candidate("HIGH_YIELD", 10.0, 0.1, True, 7.0),   # score 0.1, yield 7%
+            Candidate("MID", 10.0, 5.0, True, 3.0),
+        ]
+        # Big capital so bucket1 consumes only the top score name, leaving
+        # HIGH_YIELD (not HIGH_SCORE) to be the bucket3 top pick.
+        res = allocate_capital(1_000_000, cands)
+        b3 = [o.symbol for o in res.orders if o.bucket == 3]
+        # HIGH_YIELD (7%) should be selected in bucket3 before HIGH_SCORE (2%)
+        self.assertIn("HIGH_YIELD", b3)
+        if "HIGH_SCORE" in b3:
+            hi = b3.index("HIGH_SCORE")
+            hy = b3.index("HIGH_YIELD")
+            self.assertLess(hy, hi)
+
     def test_invalid_capital_raises(self):
         with self.assertRaises(Exception):
             allocate_capital(0, self.make_candidates())