When running the eval framework the order in which the queries are written into the final evaluation output is non-deterministic (I suspect this happens both in the VectaraConnector as well as the actual Evaluator). We should make this deterministic so the queries are always written in the same order in the output.