I would say that, in the end, the goal is to get accurate counts. One of the primary uses of the automation is so that you don’t have to do it manually, along with all of the error and bias that entails. Of course, changing your parameters could also be seen as introducing bias, but there is so much sample variation and staining variation (especially with DAB, ugh), that unless your sample prep and procedures are amazing, I don’t think you will have much choice but to deal with batch effects.
Things I would change, in order:
Preprocessing - Estimate stain vectors : This should be your bread and butter of alteration, unfortunately. Some of the main things here are alterations in the background (so make sure you have appropriate “white space” when setting this!), DAB aging/clumping, different tissue types, and often fixation uneven-ness. @Zbigniew_Mikulski gave some great information in the QuPath workshop (2019 and 2020 SD,USA) on sample preparation.
Anything below this point: Blind whomever is changing the settings. Newer versions of QuPath have the option to scramble the file names. Use it. State that you used it while adjusting the settings in any papers.
Subcellular detection - DAB Threshold : I am much less keen on changing this, as most of what you need to change should be taken care of with the Preprocessing. In the end, you may have to change it to get accurate counts, though I would definitely try to go back and re-run the new settings on old samples to see if you can make it work.
All other settings in the Subcellular detection dialog: No. Don’t touch. Or if you have to, definitely go back and rerun all samples with the same settings. Changing these would be similar to adjusting the nuclear size threshold in cell detection. Make it larger, and suddenly there are no immune cells in the whole sample! Magic!