This application scenarion, together with its use cases, comes from the goals of the Advene framework in which I've been participating.
Several users of a video library / VoD website, are allowed to annotate the videos they view or rent, and to share their annotations with other users. However, for copyright reasons, they are not allowed to modify the original video files, nor to share the files with other users. They can nevertheless refer to the URL of each video, which can in turn be viewed/rented by other users of the service.
UC1. John wants to publish comments on the last movies he has seen on http://cheap-vod.com/ . For each movie, he uses the description metadata field to provide a personal summary of the movie (with incentive to see or avoid the movie according to his own opinions), and the ranking metadata. John is also not satisfied with the genre classification of the website, so he uses the genre metadata field to provide his appreciation of the genre with regard to a better scheme. He then publishes these metadata on his blog (may be in the form of a podcast), but only links to the videos themeselves.
Jane, a friend of John's and another cheap-vod customer, can now configure her cheap-vod account or her browser, to have John's metadata added to or replacing the original metadata embeded in each file.
UC2. Jane is an english teacher. She wants to use The purple rose of Cairo with her students. The movie is available for viewing (but not downloading) at the school library. For making navigation in the movie easier, Jane identifies a fragment for each sequence, each with a title and a description. She makes that metadata available to her students on her website, with a reference to the URL of the movie at the library.
By uploading the metadata to their library account, all students can now use that metadata to enhance their vizualisation of the movie: jumping directly to a sequence, displaying the title or description of the sequence currently playing, etc.
UC3. Now Jane wants to study more particularly the characters of the movie. For making this easier, she defines one custom metadata field for each of the main characters, and sets these fields to "yes" or "no" for each sequence, to indicate if they contain that character or not. E.g. (in RDFish syntax)
<http://library.myschool.edu/rose.ogv#some_fragment_identifier> dc:title "Meeting Tom Baxter" ; dc:description "Cecilia sees the movie several times when...." ; custom:cecilia "yes" ; custom:tom "yes" ; custom:gil "no" ; custom:monk "no".
R1. (from UC1): it should be possible to have the metadata expressed in a file distinct from the video/media itself. There should be hence be means to merge metadata from several sources (intrinsic or extrinsic).
R2. (from UC2): it should be possible to annotate not only the whole media, but also fragments of the media (using the URIs to be proposed by the Media Fragment WG in order to address those fragments).
R3. (from UC3): it should be possible to extend the metadata with fields from custom ontologies.