Close
I'll go say yes to the new filenames.
Worked great on old konachan.com and when I changed to md5 due to move to danbooru was terribly welcomed.

You have the ids so finding dupes should be as easy as to order by filename

moe 8882 suzumiya_haruhi_no_yuutsu
moe 8882 the_melancholy_of_suzumiya_haruhi

They will be next to each other so it's as I said very easy.

Also it's relatively easy to calculate md5 locally on an UNIX machine, and to write a bash / sed / awk script to do it on all filenames. Actually any language can do that easy enough, java, ruby etc.

Saying MD5 are easy to find dupes is lol, if it would be really an effective way to finding dupes we wouldn't have the iqdb image similarity check on upload.

Meh