Within the last 2000 years, I believe that the published canons are probably accurate to a 5-10 minutes or so. The accuracy worsens the farther back you go because of the accuracy of the input parameters, the accumulation of numerical 'round off' errors in the computations, and of course the accumulated perturbations in the lunar orbit. I have to guess that because the Moon is only 1/2 degree across, eclipses predicted for years much earlier than, say, 3000 BC are probably hit and miss. There would be a rapid increase in false-positive predictions. Sorry, that's the best I can do on this subject.