A statistical analysis of the accuracy of the digitized magnitudes of photometric plates on the time scale of decades with an application to the century-long light curve of KIC 8462852

We present a statistical analysis of the accuracy of the digitized magnitudes of photometric plates on the time scale of decades. In our examination of archival Johnson B photometry from the Harvard DASCH archive, we find a median RMS scatter of lightcurves of order 0.15mag over the range B~9-17 for all calibrations. Slight underlying systematics (trends or flux discontinuities) are on a level of ~<=0.2mag per century (1889-1990) for the majority of constant stars. These historic data can be unambiguously used for processes that happen on scales of magnitudes, and need to be carefully examined in cases approaching the noise floor. The characterization of these limits in photometric stability may guide future studies in their use of plate archives. We explain these limitations for the example case of KIC8462852, which has been claimed to dim by 0.16mag per century, and show that this trend cannot be considered as significant.