diff --git a/quartz_solar_forecast/evaluation.py b/quartz_solar_forecast/evaluation.py
new file mode 100644
index 00000000..14a68fa2
--- /dev/null
+++ b/quartz_solar_forecast/evaluation.py
@@ -0,0 +1,34 @@
+"""
+To evaluate the performance of the solar forecast, a predefined testset is used.
+
+A file has been added to this branch (make-testset) which defines a set of random timestamps and sites ids. 
+This contains 50 sites each with 50 timestamps to make 2500 samples in total.
+
+"""
+
+def run_eval(testset_path):
+  # load testset from csv
+  testset = df.read_csv(testset_path)
+
+  # Extract generation data and metadata for specific sites and timestamps for the testset from Hugging Face.
+
+
+  # Split data into PV inputs and ground truth.
+
+
+  # Collect NWP data from Hugging Face, ICON.
+
+
+  # Run forecast with PV and NWP inputs. 
+
+
+  # Combine the forecast results with the ground truth (ts, id, horizon (in hours), pred, truth, diff)
+
+
+  # Save file
+
+
+  # Calculate and print metrics: MAE
+
+
+  # Visulisations