Experimental network research is subject to challenges when it comes to undesired effects on the experiment outcomes from the platform. In shared experiment networks the control over the network is limited, and QoS guarantees might not be available. The lack of predictability means that unwanted artifacts can be introduced in the measurements, reducing repeatability of experiments. We propose a novel, systematic, approach based on statistical analysis of metadata to deal with undesired effects on the measurements. Our proposal is a methodology where analysis of background measurements are used to identify comparable periods of network behavior. This can provide improved repeatability of an experiment and also reduce the required sample sizes. We present a proof-of-concept implementation of our method, deployed in the FEDERICA network.