Musescore.com pdf import seems to have gone downhill.

I remember getting reasonable output from musescore.com/import most of the times I used it last year. Occasionally useless, and never perfect, but usually close enough that copying things from the output into a new score and fixing the problems would be quite a bit faster than retypesetting.

But on the few occasions I've used it this year, it's given me absolute garbage. Opening these broken files in Musescore tells me they're 'corrupted', and "show details" gives a long list of errors like this:

"Measure 12, staff 2 incomplete. Expected: 47/32; Found: 218/96"

If I say to ignore the errors and open anyways, working with these files often results in Musescore crashing. And even if it weren't crashing &c, the completely wrong key signatures, bizarre 64th notes, random rests inserted between notes, etc would be many times more work to correct than simply re-typesetting the piece from scratch.

I tried installing the latest release of audiveris myself. Out of the box a primary problem was immediately evident: a lot of musical elements didn't find their way from the pdf to the binarized image Audiveris was working on. Switching from adaptive to global binarization and adjusting the threshold gave me results as good as those I remember getting from musescore.com last year.

If I remember correctly, the importer used to also output MusicXML (as Audiveris does) rather than mscz. Is it possible that some problems are creeping up in the conversion phase? It'd be interesting to see what the score looks like at different stages of the musescore.com/import - the binarized image, the audiveris book, audiveris xml out, then the mscz.

Two other import related thoughts: Data retention is a problem. You need to give people a way to remove their old scores if they don't want you storing them forever. And you've misspelt mscz as mcsz in 'This is a free experimental service, that attempts to help you out to convert your PDF sheet music to a .mcsz file.'