Well, it’s tough being as badly calibrated as the Vaillant sensors. Any DS18B20 is typically closer to the truth than those particular NTC10K.
You’ve got a baseline 0.6K delta, which means you overestimate COP. The amount depends on your actual deltaT when the compressor is on and ideltaT/(deltaT-0.6K) will give you the “wrongness factor”. For example, 1.3 would indicate 30% overestimation.