The difference between a Sensor and a Stream in the above example is that in the case of the sensor the user is primarily concerned with the actual value of the "Value". Meaning a spike in temperature or a drop in CPUs. But in a stream of customer data there isn't a time "X" and value "Y" there are many values "Y" and you a user is interested in the overall quality of both the entire stream and the individual values. Relationship analysis and other correlative functions apply here. If you were to chart a "stream" what would you chart? The row count volume or just one of the columns or the count of something? But if you were to chart a sensor you know exactly what you would chart... the "Value" over "Time".
Fortunately Owl has already thought and worked through the many nuances required to understand, monitor and predict accurately for all of these use-case. All that is required is to subscribe the stream.