feat: enrich training deploy via mimir sdg

tests/test_generate_training_data.py

@@ -49,6 +49,7 @@ class GenerateTrainingDataTests(unittest.TestCase):
             "sdg_host": None,
             "sdg_user": None,
             "sdg_key": None,
+            "sdg_model_alias": "nvidia-text",
             "num_records": 1000,
             "timestamp": 1234567890,
         }
@@ -120,6 +121,58 @@ class GenerateTrainingDataTests(unittest.TestCase):
             "training-data/module_controller_intents_123",
         )
 
+    def test_data_designer_config_declares_structured_expansion(self):
+        config = bridge.data_designer_config(
+            dataset_key="module_controller_intents_123",
+            generated_at=123,
+            remote_seed="/tmp/seed.jsonl",
+            action_ids=["logger"],
+            model_alias="nvidia-text",
+        )
+        self.assertIn("load_config_builder", config)
+        self.assertIn('column_type="llm-structured"', config)
+        self.assertIn('model_alias="nvidia-text"', config)
+        self.assertIn("expanded_record", config)
+
+    def test_remote_data_designer_command_sources_mimir_env(self):
+        command = bridge.remote_data_designer_create_command(
+            remote_config="/tmp/controller.py",
+            num_records=42,
+            dataset_key="module_controller_intents_123",
+            artifact_path="/tmp/artifacts",
+        )
+        self.assertIn(bridge.MIMIR_DATA_DESIGNER_ENV, command)
+        self.assertIn(bridge.MIMIR_DATA_DESIGNER_BIN, command)
+        self.assertIn("--num-records 42", command)
+        self.assertIn("--no-tui", command)
+
+    def test_remote_bash_command_quotes_inner_command(self):
+        command = bridge.remote_bash_command("set -a; echo ready")
+        self.assertTrue(command.startswith("bash -lc "))
+        self.assertIn("'set -a; echo ready'", command)
+
+    def test_normalize_expanded_jsonl_extracts_expanded_record(self):
+        raw = self.root / "expanded.raw.jsonl"
+        normalized = self.root / "expanded.jsonl"
+        source_row = {
+            "expanded_record": {
+                "instruction": bridge.INSTRUCTION,
+                "input": "record this message",
+                "output": json.dumps({"action_id": "logger", "payload": {"message": "hello"}}),
+                "metadata": {
+                    "action_id": "logger",
+                    "source": "data_designer",
+                    "split": "train",
+                    "generated_at": 123,
+                },
+            }
+        }
+        bridge.write_jsonl(raw, [source_row])
+        bridge.normalize_expanded_jsonl(raw, normalized)
+        rows = bridge.read_jsonl(normalized)
+        self.assertEqual(len(rows), 1)
+        self.assertEqual(rows[0]["metadata"]["action_id"], "logger")
+
 
 if __name__ == "__main__":
     unittest.main()