The rules state a maximum of 1,000 training samples per pilot region including self-collected samples, but the Full Rules also say to use only the provided datasets. Can you confirm whether we may add our own labeled samples (using freely accessible Sentinel data) up to the 1,000 per region cap?