diff --git a/.github/workflows/test-all.yaml b/.github/workflows/test-all.yaml
index 9fd9a84..f3df270 100644
--- a/.github/workflows/test-all.yaml
+++ b/.github/workflows/test-all.yaml
@@ -13,6 +13,14 @@ jobs:
       - name: Test Python 3.9
         run: make test_3_9
 
+  test-py-3-9-validation:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Test Python 3.9 (validation)
+        run: make test_validation_3_9
+
   test-py-3-10:
     runs-on: ubuntu-latest
     steps:
@@ -21,6 +29,14 @@ jobs:
       - name: Test Python 3.10
         run: make test_3_10
 
+  test-py-3-10-validation:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Test Python 3.10 (validation)
+        run: make test_validation_3_10
+
   test-py-3-11:
     runs-on: ubuntu-latest
     steps:
@@ -28,3 +44,11 @@ jobs:
 
       - name: Test Python 3.11
         run: make test_3_11
+
+  test-py-3-11-validation:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+
+      - name: Test Python 3.11 (validation)
+        run: make test_validation_3_11
diff --git a/Makefile b/Makefile
index 78edb67..db9039a 100644
--- a/Makefile
+++ b/Makefile
@@ -44,6 +44,14 @@ test_3_11: build_3_11 ## Test Python 3.11 pickle
 	docker run -i --rm -v ${PWD}:${PWD} -v /tmp:/tmp ${NAME}_py_3_11:latest --model ${PWD}/tests/models/model_3_11_legacy.pkl
 	docker run -i --rm -v ${PWD}:${PWD} -v /tmp:/tmp ${NAME}_py_3_11:latest --model ${PWD}/tests/models/model_3_11.pkl
 
+test_validation_%: build_% ## Test validation dataset
+	docker run -i --rm -v ${PWD}:${PWD} -v /tmp:/tmp ${NAME}_py_$*:latest \
+		--dataset v4.3/validation_int8.parquet --benchmarks v4.3/validation_benchmark_models.parquet \
+		--model ${PWD}/tests/models/model_$*_legacy.pkl
+
+.PHONY: test_validation
+test_validation: test_validation_3_9 test_validation_3_10 test_validation_3_11
+
 .PHONY: push_latest
 push_latest: push_latest_3_9 push_latest_3_10 push_latest_3_11 ## Push latest docker containers
 
diff --git a/predict.py b/predict.py
index bd7529f..a93fe1e 100644
--- a/predict.py
+++ b/predict.py
@@ -183,37 +183,53 @@ def main(args):
     if num_args > 1:
         benchmark_models = get_data(args.benchmarks, args.output_dir)
 
-    logging.info(f"Predicting on {len(live_features)} rows of live features")
+    num_eras = live_features["era"].nunique()
+
+    if num_eras > 1:
+        logging.info(
+            f"Predicting on {len(live_features)} rows, {num_eras} eras of features"
+        )
+    else:
+        logging.info(f"Predicting on {len(live_features)} rows of live features")
+
     try:
-        if num_args == 1:
-            predictions = model(live_features)
-        elif num_args == 2:
-            predictions = model(live_features, benchmark_models)
-        else:
-            logging.error(
-                f"Invalid pickle function - {model_pkl} must have 1 or 2 arguments"
-            )
-            exit_with_help(1)
-
-        if predictions is None:
-            logging.error("Pickle function is invalid - returned None")
-            exit_with_help(1)
-        elif type(predictions) != pd.DataFrame:
-            logging.error(
-                f"Pickle function is invalid - returned {type(predictions)} instead of pd.DataFrame"
-            )
-            exit_with_help(1)
-        elif len(predictions) == 0:
-            logging.error("Pickle function returned 0 predictions")
-            exit_with_help(1)
-        elif predictions.isna().any().any():
-            logging.error("Pickle function returned at least 1 NaN prediction")
-            exit_with_help(1)
-        elif not (predictions.iloc[:, 0].between(0, 1).all().all()):
-            logging.error(
-                "Pickle function returned invalid predictions. Ensure values are between 0 and 1."
-            )
-            exit_with_help(1)
+        predictions = []
+        for era, era_features in live_features.groupby("era"):
+            if num_eras > 1:
+                logging.debug(f"Predicting era {era} with {len(era_features)} rows")
+            if num_args == 1:
+                era_predictions = model(era_features)
+            elif num_args == 2:
+                era_benchmark_models = benchmark_models.loc[era_features.index]
+                era_predictions = model(era_features, era_benchmark_models)
+            else:
+                logging.error(
+                    f"Invalid pickle function - {model_pkl} must have 1 or 2 arguments"
+                )
+                exit_with_help(1)
+            if era_predictions is None:
+                logging.error("Pickle function is invalid - returned None")
+                exit_with_help(1)
+            elif type(era_predictions) != pd.DataFrame:
+                logging.error(
+                    f"Pickle function is invalid - returned {type(era_predictions)} instead of pd.DataFrame"
+                )
+                exit_with_help(1)
+            elif len(era_predictions) != len(era_features):
+                logging.error(
+                    f"Pickle function returned {len(era_predictions)} predictions, expected {len(era_features)}"
+                )
+                exit_with_help(1)
+            elif era_predictions.isna().any().any():
+                logging.error("Pickle function returned at least 1 NaN prediction")
+                exit_with_help(1)
+            elif not (era_predictions.iloc[:, 0].between(0, 1).all().all()):
+                logging.error(
+                    "Pickle function returned invalid predictions. Ensure values are between 0 and 1."
+                )
+                exit_with_help(1)
+            predictions.append(era_predictions)
+        predictions = pd.concat(predictions)
     except TypeError as e:
         logging.error(f"Pickle function is invalid - {e}")
         if args.debug: