diff --git a/plexe/retrain.py b/plexe/retrain.py
index 18d7dd50..8c1eedf5 100644
--- a/plexe/retrain.py
+++ b/plexe/retrain.py
@@ -419,7 +419,7 @@ def reset_weights(module):
             metric_name = "unknown"
 
         solution = Solution(
-            iteration=0,
+            solution_id=0,
             feature_pipeline=fitted_pipeline,
             model=untrained_model,
             model_type=model_type,
diff --git a/tests/unit/test_retrain.py b/tests/unit/test_retrain.py
new file mode 100644
index 00000000..6e67f9c1
--- /dev/null
+++ b/tests/unit/test_retrain.py
@@ -0,0 +1,79 @@
+"""Tests for plexe.retrain.retrain_model using a fabricated xgboost model package."""
+
+import json
+import tarfile
+from unittest.mock import MagicMock
+
+import joblib
+import pandas as pd
+import pytest
+
+from plexe.models import Solution
+from plexe.retrain import retrain_model
+
+
+@pytest.fixture
+def retraining_inputs(tmp_path):
+    """Create a minimal retraining-ready model.tar.gz plus matching dataset."""
+    pkg = tmp_path / "pkg"
+    (pkg / "artifacts").mkdir(parents=True)
+    (pkg / "src").mkdir(parents=True)
+
+    df = pd.DataFrame(
+        {
+            "x1": range(40),
+            "x2": [(i * 7) % 13 for i in range(40)],
+            "y": [(i * 3) % 5 for i in range(40)],
+        }
+    )
+
+    from xgboost import XGBRegressor
+
+    model = XGBRegressor(n_estimators=2, max_depth=1)
+    model.fit(df[["x1", "x2"]], df["y"])
+    joblib.dump(model, pkg / "artifacts" / "model.pkl")
+
+    metadata = {"model_type": "xgboost", "target_column": "y", "task_type": "regression", "experiment_id": "orig"}
+    (pkg / "artifacts" / "metadata.json").write_text(json.dumps(metadata))
+    (pkg / "src" / "trainer.py").write_text("# trainer\n")
+    (pkg / "src" / "pipeline.py").write_text(
+        "from sklearn.pipeline import Pipeline\n"
+        "from sklearn.preprocessing import StandardScaler\n"
+        "pipeline = Pipeline([('scaler', StandardScaler())])\n"
+    )
+
+    tarball = tmp_path / "model.tar.gz"
+    with tarfile.open(tarball, "w:gz") as tar:
+        for item in pkg.rglob("*"):
+            if item.is_file():
+                tar.add(item, arcname=item.relative_to(pkg).as_posix())
+
+    dataset = tmp_path / "new_data.parquet"
+    df.to_parquet(dataset, index=False)
+    return tarball, dataset
+
+
+def test_retrain_model_returns_solution(retraining_inputs, tmp_path):
+    """retrain_model must complete and return a valid Solution after training."""
+    tarball, dataset = retraining_inputs
+
+    artifacts_dir = tmp_path / "training_run" / "model_artifacts"
+    (artifacts_dir / "artifacts").mkdir(parents=True)
+    runner = MagicMock()
+    runner.run_training.return_value = artifacts_dir
+    config = MagicMock()
+    config.training_timeout = 60
+
+    solution, metrics = retrain_model(
+        original_model_uri=str(tarball),
+        train_dataset_uri=str(dataset),
+        experiment_id="exp-retrain",
+        work_dir=tmp_path / "work",
+        runner=runner,
+        config=config,
+    )
+
+    assert isinstance(solution, Solution)
+    assert solution.model_type == "xgboost"
+    assert metrics["metric"] == "unknown"  # fabricated package intentionally has no model.yaml
+    assert (tmp_path / "work" / "model.tar.gz").exists()