To a Roon analyst/designer
In this case ROON groups correctly ALL the compositions of an Album.
But it fails to display the correct name for some the compositions.
This is caused by the use of “Op.”
Roon documentation seems to be silent about the usage of “Op.”
An issue for the maintenance team.
Perhaps someone from roon will comment but Opus numbers (and/or catalog numbers) plus a composer are pretty much essential to a correct composition identification with an unidentified album as roon will not have much else to go on. So sounds like there is something else going on.
Can you make a screenshot of something incorrectly named and what you expected to see?
PS. I could only find one Arrau Beethoven Complete Sonatas. Is it this one or did you make a virtual disk from various albums?
This may have been said above, but if so it bears repeating… It is folly and unrealistic to expect Roon (v1.5) to correctly parse and group classical works.
I learned this lesson the hard way. I would dump classical stuff into Roon’s watch folder “unvarnished”: no regard to tag contents, no pre-tagging, nothing. And I got what I deserved. It was a hit-and-miss hodgepodge of results. Some classical albums worked fine, but I would put their percentage at under 50 percent of the total. More than half were incompletely displayed in one way or another: works ungrouped and/or un-IDed; composers missing or appearing as “artists”; etc.
To increase your odds to nearly 100 percent, buy a $12 Allmusic account, buy a good tagger (my favorite Mac app is Yate), find the exact verbiage Allmusic uses for Works (lesser importance for Parts), and paste it into each track’s WORK tag for all its tracks. While you’re there, do the same thing for the PART if you wish. Try to do ALL major editing with a tagger, not Roon. Why? Transparency. A tagger will easily show you what is affecting the appearance.
You will often scratch your head over how often just a little variance in a work’s title will cause Roon to miss the ID. For example, your pre-populated tags may have, “Sonata in D major, Op. 22” and will miss the id until you enter, “Sonata for piano in D major, Op. 22” Or, the tag may contain a work nickname, eg “Titan”, that is not enclosed in parentheses.
I used to lookup names on the IMSLP website, or on whatever showed up at the top of the Google search. I quickly found that ONLY Allmusic’s canonical name works reliably. Using other title versions is akin to fighting city hall.
Another misstep I made was relying on the titles displayed in Roon as good ones to use editing un-IDed works. NOT SO. Roon uses the TITLE tag to display a composition name, but uses WORK and PART tags (where available) to actually ID the composition.
I used to believe that if an album had been IDed correctly, then so too would be all the works therein. Not necessarily so. I discovered dozens of albums “correctly” IDed but without a single work being IDed! And it is no longer that easy to detect whether a work is IDed. For example, I had, at one time, nine "Bolero"s that had the the little album icon next to them. But clicking on the icon revealed that Roon had linked all of them to each other but none of them to the rich metadata. Only by changing the tags to the canonical names a la Allmusic would they link.
You cannot rely on your eyes and common sense to discern whether a composition has been correctly tagged. Dvorak’s work needs B numbers instead of opus numbers. To keep opuses, one puts them in parentheses. If you own a Debussy album cut between 1977 and 2003, you cannot rely on the album cover for the correct description. The L numbers were amended in 2003.
Even after editing the canonical names, Roon may still refuse to ID unless it has a known “exemplar” to compare against, typically a TIDAL import. I have literally hundreds of TIDAL album imports whose only purpose in my library is to induce identification.
So, either invest the time and tag your stuff correctly, or take whatever you may get from Roon. It does little good to wonder why Roon does what it does, and it wastes time. Roon doesn’t answer “why” as a rule.
Good luck to you.
John
I fully agree with all you said, as it matches 100% my experience over many years. In the current state of affairs, the best we can do is to wait for Roon to deliver on their promise of an improved composition identification mechanism, where Rovi inedaquacies can be bypassed, and user choices and decisions can be acknowledged. But how long will it be before this occurs ?
Is it still useful to keep reporting on Roon composition identification problems ? I have stopped
doing it for a long time now, as I feel it is a waste of time without an active Roon participation in the discussion and without a way to make knowledgeable users like you and many others active contributors to a solution. I am strongly convinced that Roon will not make it without involving its user base.
Regards
And I, in turn, agree with you, Andre. But just remember, we classical types are only 5% of the audience, so we can be safely unsatisfied.
Then again, losing 5% can be the difference between a profit and a loss. hmm…
A new and final development for me.
I copied the sonata names from this album on All Music:
Beethoven: The Complete Piano Sonatas & Concertos
Claudio Arrau
1998
Having copied the sonata names as values for my offending sonatas I had no issue any more.
I understand the rational to try to stick to an existing database.
For me this is fine but it is not a priority.
I can now see the proper sonata names and have avoided to have to live with the wrong names picked by Roon.
That was the target.
In the future I will remember John Venable. Logic does not pay.
I’m gonna have to save this quote! 
My own Songkong does automatically add WORK and PART fields for classical albums when that info is available within Musicbrainz db or can be derived from track titles. So that should make things easier for you