Ran 4 open-source geo-experiment estimators on 8,000 synthetic panels with planted ground truth. Their point estimates look interchangeable, but their uncertainty isn't.