I come to realise, today it is the (nominal) 2500th anniversary since the famous historic Xerxes' solar eclipse on 2nd October 480BC... why is that of relevance in a Celestia forum? if you recall, we used that as the reference to help out improving the accuracy of Celestia for version 1.2.5 (Jan 2003) - for me, that bug report has been my all time favourite! All we had to do, was just to pass along the right image:
Celestia 1.2.5 8 Jan 2003
Release 1.2.5 of Celestia is finally ready. The new release includes a lot of bug fixes and some major new features. Linux users should be pleased with the new KDE interface (though the Gnome interface is still there, and even has bug fixes.) In 1.2.5 comet tails are rendered, so you can see striking views of Hale-Bopp in 1997 and other famous comet encounters. The accuracy of Celestia has improved dramatically. It's possible now to watch the moons of Jupiter eclipse each other, follow the Voyager spacecraft on their grand tours of the solar system, and see the eclipse of Xerxes in 479 B.C.E. The complete list of changes is in the change log.
So, I did the usual trick and went to see how it compares, my impression is that it is pretty near, however it is not spot on. Practically, any pair of eclipses, which are set several thousands of years apart, would serve the purpose of confining the errors within the fully contained Saros cycles. So, I think this is a worthy exercise in its own right. Admittedly, I have not tried to compare with the JPL ephemerides.
Q. Anyone care to comment? There is also an open github issue about the topic, #23. Hint: enable eclipse shadows and use this Celestia link:
For eclipse aficionados Ref. https://eclipse.gsfc.nasa.gov/SEhistory/SEhistory.html