Digital sensors are used in the sensors, ie not resistance measuring elements, but elements with a communication bus, which already give a numerical value of the measured temperature (or humidity). The measurement error of the element itself may be different from that stated in the catalog sheet of their manufacturer and is therefore practically ruled out.
This correction is already set at the factory and we set it so that it maximally compensates for the actual inaccuracy of the measuring element. However, as the practice shows, an increase in the measured temperature can occur even if the supply voltage of the controller exceeds 27 V AC. Then they take over the overvoltage protections, cut off the tips of the power sine wave, and convert this energy into heat, which then affects the measured temperature. Check this first. Today's transformers can have a no-load voltage slightly higher than at full load (when it should not fall below the nominal 24 V AC), and therefore this phenomenon can occur at low transformer loads. The countermeasure is to use a different power supply for the drivers.
The customer usually uses a laboratory mercury thermometer or an electronic probe as a reference thermometer. Both of these devices have a measuring element of small dimensions and usually glossy, ie only the temperature of the medium (air) is measured, this thermometer is intended for this, not the radiant component. However, after impact on the plastic cover of the sensor, it is sensed and affects the measured temperature. The controllers therefore measure not only the air temperature, but also the radiant component (this is perfectly fine and desirable from the point of view of regulation, because this is exactly how a person perceives temperature). The contribution of the radiant component can be about 2 K.
We also found that the customer used a cheap digital thermometer located on a coffee table, about 3 m from the sensor and 1 m lower. A view with a thermal imager proves that the temperature conditions in the room can have a large variance.