Another recommendation would be to unlock the ability for a customer to use the previous version of the software
This would have been done, if it were possible. It was not our wish to force updating to the latest version, but the need. Due to some third party changes, versions 850 and bellow started providing totally inadequate results (very poor detection), thus we were to force-update our clients.
T-comparator presets are not so easily changed. Presets currently used are based on the algorithms developed for PAN'14 conference, plagiarism detection task, and showed one of the best results with the used document corpus. One of the lines of our research now is what is different between that corpus and the "real-life" cases, which can be closely connected to the problem you report.
Changing the preset in the current system requires some research, as we need to be sure the new preset works better. At present we have started additional tests on the false-positive detection and upon the results we will both post here and may start making the needed alterations.