diff --git a/src/evalopt_graph/kernel.py b/src/evalopt_graph/kernel.py
index cc18763..e4f1dd8 100644
--- a/src/evalopt_graph/kernel.py
+++ b/src/evalopt_graph/kernel.py
@@ -157,10 +157,23 @@ def __post_init__(self) -> None:
             raise ValueError("observed_at must be a string")
 
         def rows(name: str, values: Iterable[Iterable[str]], width: int) -> tuple[tuple[str, ...], ...]:
-            normalized = tuple(tuple(row) for row in values)
+            if isinstance(values, str | bytes):
+                raise ValueError(f"{name} entries require {width} string fields")
+            normalized: list[tuple[str, ...]] = []
+            try:
+                iterator = iter(values)
+            except TypeError as exc:
+                raise ValueError(f"{name} entries require {width} string fields") from exc
+            for row in iterator:
+                if isinstance(row, str | bytes):
+                    raise ValueError(f"{name} entries require {width} string fields")
+                try:
+                    normalized.append(tuple(row))
+                except TypeError as exc:
+                    raise ValueError(f"{name} entries require {width} string fields") from exc
             if any(len(row) != width or not all(isinstance(item, str) for item in row) for row in normalized):
                 raise ValueError(f"{name} entries require {width} string fields")
-            return normalized
+            return tuple(normalized)
 
         object.__setattr__(self, "gate_results", rows("gate_results", self.gate_results, 2))
         object.__setattr__(self, "contradictions", rows("contradictions", self.contradictions, 3))
@@ -217,10 +230,10 @@ def to_dict(self) -> dict[str, Any]:
     def from_dict(cls, value: Mapping[str, Any]) -> AcceptanceInput:
         return cls(
             observed_at=str(value.get("observed_at", "")),
-            gate_results=tuple(tuple(item) for item in value.get("gate_results", ())),
+            gate_results=value.get("gate_results", ()),
             criteria=tuple(value.get("criteria", ())),
             claims=tuple(ClaimRecord(**item) for item in value.get("claims", ())),
-            contradictions=tuple(tuple(item) for item in value.get("contradictions", ())),
+            contradictions=value.get("contradictions", ()),
             attestations=tuple(EvidenceAttestation.from_dict(item) for item in value.get("attestations", ())),
             assessments=tuple(SupportAssessment.from_dict(item) for item in value.get("assessments", ())),
             tests_weakened=value.get("tests_weakened", False),
diff --git a/tests/test_kernel.py b/tests/test_kernel.py
index 524ec1d..7b91617 100644
--- a/tests/test_kernel.py
+++ b/tests/test_kernel.py
@@ -245,6 +245,51 @@ def test_nested_input_rows_are_copied_before_evaluation():
     assert before.status == "ACCEPTED"
 
 
+@pytest.mark.parametrize(
+    ("field", "bad_row"),
+    [
+        ("gate_results", "ok"),
+        ("gate_results", b"ok"),
+        ("contradictions", "abc"),
+        ("contradictions", b"abc"),
+    ],
+)
+def test_acceptance_input_rows_reject_scalar_rows_instead_of_splitting_characters(field, bad_row):
+    with pytest.raises(ValueError, match=f"{field} entries"):
+        kernel.AcceptanceInput(observed_at=NOW, **{field: (bad_row,)})
+
+    with pytest.raises(ValueError, match=f"{field} entries"):
+        kernel.AcceptanceInput.from_dict(
+            {
+                "schema_version": "evalopt.acceptance-input.v1",
+                "observed_at": NOW,
+                "gate_results": [],
+                "criteria": [],
+                "claims": [],
+                "contradictions": [],
+                "attestations": [],
+                "assessments": [],
+                "tests_weakened": False,
+                "evaluator_score": None,
+                field: [bad_row],
+            }
+        )
+
+
+@pytest.mark.parametrize(
+    ("field", "bad_value"),
+    [
+        ("gate_results", None),
+        ("gate_results", 1),
+        ("contradictions", None),
+        ("contradictions", 1),
+    ],
+)
+def test_acceptance_input_rows_reject_non_iterable_row_containers(field, bad_value):
+    with pytest.raises(ValueError, match=f"{field} entries"):
+        kernel.AcceptanceInput(observed_at=NOW, **{field: bad_value})
+
+
 def test_contradiction_enums_are_normalized_and_unknown_values_rejected():
     policy, input_ = _input(contradictions=(("x1", "HIGH", "RESOLVED"),))