As Martin noted the magnitude params computed by the MPC are based mostly on estimates made as a byproduct of doing astrometry. They are therefore generally based on nuclear magnitudes which will be underestimates for comets with larger apparent diameters. The COBS data are generally total magnitudes and the magnitude parameters derived from those are generally better than the MPC ones assuming that there are enough observations to work with. Jonathan uses total magnitudes too and adds a human element and much experience.
The magnitude of a comet depends on its activity level and this can be quite unpredictable. Even well-known periodic comets can behave badly (cf 17P/Holmes). That is what makes them fun to observe! You identified a good example in 289P/Blanpain which was mentioned in the 2019 Handbook with a big warning about its magnitude:
C/2017 T2 is another example where the mag params are not well known at the moment:
Again, this is why we are observing it.
I guess that the key message is that no fully automated comet prediction website is likely to work. There will always need to be a human element checking what it says and adapting it to the latest observations.