Sensors may report real numbers or conditional states (such as on/off or raining/not raining). Even if it is reporting conditional states, in many cases, the signal is continuous and some logic determines if it has crossed a threshold that is interpreted as a change in state.

Sensors vary in accuracy, and there is usually a cost trade off in order to increase measurement accuracy.

External conditions may also affect accuracy. For example, extremes in cold can affect the accuracy of some motion sensors. This is important to know for analytics, as it can affect the results of prediction from machine-learning models. This external influencer can be what is referred to as a confounding variable. It has an effect on two or more measured variables but is not itself directly measured.

You may need to adjust for it when processing the data for analytics on the backend. In the case of the temperature example, you can apply a formula to adjust the reading based on external weather data you will have mashed in on the backend.

There is also some level of noise in sensor readings that sometimes has to be filtered or transformed to smooth out the reported values. This often happens on the device itself through various algorithms. The algorithms employed on the device to infer measurements may be implemented incorrectly resulting in some misreading of values.

Most of these issues will be caught and corrected by the product validation processes. However, it is important to know if there are any product limitations or adjustments needed when it is near the edges of its operating ranges.