diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9632571d..c479367b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -47,6 +47,7 @@ All notable changes to OriginWeave are documented in this file. The format follo
 ### Security
 
 - Raw page content cannot become a trusted instruction.
+- Controlled Agent Task compatibility evidence is machine-checked to contain only the reviewed `input` and `submit` browser-computed role/name fields before bounded measurement, preventing unreviewed page text or instruction-like fields from silently entering that evidence object.
 - Raw secrets are rejected and secret-capable actions require an opaque broker handle.
 - Crawler mode is read-only, must pair with the public-crawl purpose, and fails closed without an applicable robots-policy decision.
 - State-changing actions are same-origin by default.
@@ -75,4 +76,4 @@ All notable changes to OriginWeave are documented in this file. The format follo
 - The hourly product agent has no Git metadata or repository authority. A separate post-verification publisher opens one PR and cannot approve or merge it.
 - The unprivileged OpenCode user is restricted to loopback egress during model execution, preventing runner-wide allow-listed endpoints from becoming direct source-exfiltration channels.
 
-[Unreleased]: https://github.com/ContextualWisdomLab/OriginWeave/compare/main...HEAD
\ No newline at end of file
+[Unreleased]: https://github.com/ContextualWisdomLab/OriginWeave/compare/main...HEAD
diff --git a/tests/test_agent_task_semantic_observation_schema_contract.py b/tests/test_agent_task_semantic_observation_schema_contract.py
new file mode 100644
index 00000000..87b9112e
--- /dev/null
+++ b/tests/test_agent_task_semantic_observation_schema_contract.py
@@ -0,0 +1,90 @@
+"""Contract for the controlled Agent Task semantic-observation evidence schema."""
+
+from __future__ import annotations
+
+import ast
+import pathlib
+import unittest
+
+ROOT = pathlib.Path(__file__).resolve().parents[1]
+RUNNER = ROOT / "scripts" / "ci" / "run_mv3_compatibility.py"
+
+
+class AgentTaskSemanticObservationSchemaContractTests(unittest.TestCase):
+    """Keep unreviewed page content outside the controlled semantic evidence object."""
+
+    @classmethod
+    def setUpClass(cls) -> None:
+        cls.tree = ast.parse(RUNNER.read_text(encoding="utf-8"), filename=str(RUNNER))
+
+    @staticmethod
+    def _literal_dict_keys(node: ast.Dict) -> tuple[str, ...]:
+        """Return exact string-literal dictionary keys or fail the contract."""
+
+        keys: list[str] = []
+        for key in node.keys:
+            if not isinstance(key, ast.Constant) or not isinstance(key.value, str):
+                raise AssertionError("semantic observation keys must be string literals")
+            keys.append(key.value)
+        return tuple(keys)
+
+    def _semantic_observation_assignment(self) -> ast.Dict:
+        """Find the one executable controlled-observation construction."""
+
+        assignments = [
+            node
+            for node in ast.walk(self.tree)
+            if isinstance(node, ast.Assign)
+            and any(
+                isinstance(target, ast.Name) and target.id == "semantic_observation"
+                for target in node.targets
+            )
+        ]
+        self.assertEqual(len(assignments), 1)
+        value = assignments[0].value
+        self.assertIsInstance(value, ast.Dict)
+        return value
+
+    def test_controlled_observation_has_only_reviewed_role_name_fields(self) -> None:
+        """Raw page text or instruction-like fields cannot drift into emitted evidence."""
+
+        observation = self._semantic_observation_assignment()
+        self.assertEqual(self._literal_dict_keys(observation), ("input", "submit"))
+        semantic_nodes = dict(zip(self._literal_dict_keys(observation), observation.values))
+
+        expected_values = {
+            "input": {"role": "input_role", "name": "input_name"},
+            "submit": {"role": "submit_role", "name": "submit_name"},
+        }
+        for semantic_key, expected_fields in expected_values.items():
+            semantic_node = semantic_nodes[semantic_key]
+            self.assertIsInstance(semantic_node, ast.Dict)
+            self.assertEqual(self._literal_dict_keys(semantic_node), ("role", "name"))
+            actual_fields = dict(
+                zip(self._literal_dict_keys(semantic_node), semantic_node.values)
+            )
+            for field_name, expected_variable in expected_fields.items():
+                value = actual_fields[field_name]
+                self.assertIsInstance(value, ast.Name)
+                self.assertEqual(value.id, expected_variable)
+
+    def test_exact_observation_flows_through_bounded_measurement(self) -> None:
+        """The reviewed object must be the exact object sent to the byte-bound helper."""
+
+        calls = [
+            node
+            for node in ast.walk(self.tree)
+            if isinstance(node, ast.Call)
+            and isinstance(node.func, ast.Name)
+            and node.func.id == "_measure_agent_task_semantic_observation_bytes"
+        ]
+        self.assertEqual(len(calls), 1)
+        self.assertEqual(len(calls[0].args), 1)
+        argument = calls[0].args[0]
+        self.assertIsInstance(argument, ast.Name)
+        self.assertEqual(argument.id, "semantic_observation")
+        self.assertFalse(calls[0].keywords)
+
+
+if __name__ == "__main__":
+    unittest.main()