The continuous streaming of data from a sensor network is essential for a digital twin to mirror its physical counterpart in real-time. Missing or interrupted data limits the digital twin functionality but can be difficult to detect when data is communicated at a change of value. A disruption in data streaming can compromise building surveillance or lead to overlooked critical alerts. Although setting basic thresholds can sometimes be effective, a more sophisticated approach requires monitoring data counts using statistical methods. By understanding the unique streaming patterns of each controller, these models are less susceptible to errors, reducing the incidence of false alarms, ensuring more accurate streaming, and notifying users immediately about any significant deviations from the expected data counts. This paper presents the development of such an approach implemented for a large multi-use building, which consists of 14,000+ points that report whenever a change of value above a defined threshold is observed. Before transmission to cloud services, streaming software structures the records with appropriate formatting and archives it in a timeseries database for subsequent analysis. The streaming quality model functions in a two steps. First, an appropriate model is determined for each controller by fitting the count distribution with known statistical distributions and an Expectation-Maximization algorithm determines their parameters. Second, the calibrated models assess on an hourly basis whether the count aligns with the expected range. If deviations exceed a set threshold, an alert is triggered. A case study implementation of this approach for a full-building digital twin is presented, showcasing both the need for and value of this approach.