Reference management systems

Writing up the dissertation is nice. That's if you're an asocial duteous fan of your ideas and experiments. I like doing everything else as well, not because of the lack of commitment, but there needs to be balance in order to understand and repeat to yourself that what you do, is useful. I must admit, however, that for the past 6 months I may have become that asocial duteous fan of the experiments.

I think I am not the first or the last to notice that cleaning up references is a true nuisance. And that's even the case if you are like me and manage most of your references in a nice structurized database. Because there's the 5% of references + other double checking that can totally enrage you. So, here's what I want. I need a collaborative reference management system, with the following properties:

* Collaborative system and online and opt-in/opt-out privacy protection for the lists - I'm not interesting in sharing the references WHILE analyzing them and working them through - you are welcome to read my paper as a tech report before it gets published, but not the raw work-in-progress bibliography;

* It has to support several statuses for the entries, comments and opt-in sharing with limited set of users + article attachment and supplementary material should be possible;

* Some intelligent entity recognition is needed really bad - you are not showing your list to others, but using invisibly other references for the same entry could collaboratively allow quick clean-up. Google has a preliminary unique ID for all papers in Scholar, but it's out of reach through API or directly from the query result;

* Support for the import and (rich text) export to most of the reference styles (APA, MLA, IEEE etc.) and bibliogrpahic formats (bibtex).

Is that too much to ask?

If there is something like that out there, please, let me know!
(No, Bibsonomy is still far from those goals!)

Oh well, back to work with essentially redundant and unproductive clean-up of the references ...