Dear Beki,
thanks for posting! Short answer is no, I'm afraid we don't have a way to "mark as excluded" or assign codes to references while working on the duplicates window. However, you are suggesting a very useful feature, and you have got me and my colleagues seriously thinking about how to implement this.
It will not be easy, as there are numerous implications that should be taken into account (not to mention the need to offer a clear and ergonomic user interface). I am not saying that this feature will not enter into our roadmap; all I'm saying is that I can't make any promises. However, I have in mind a few (pretty urgent) additions to the duplicate checking features, so I will discuss your ideas with the rest of the team and try to agree how to implement it; with a little luck, we might be able to write it along with the other de-dup new features.
Unfortunately, I don't expect this to be ready in days, so I'm afraid it might not be published in time for you to take advantage of it.
Back on the subject, it is true that 'automatically marking' duplicates with a threshold lower than 1 is dangerous, however, it might be worth taking a calculated risk (especially for reviews that contain 10'000+ references): if you have already manually evaluated at least a hundred groups, you may have the possibility to indentify a 'safe' threshold to use with the 'advanced mark automatically' function. Take a look at the 'completed' groups, and find the highest similarity score applied to items that you have identified as 'not a duplicate' (I'll call this value T). Let's say that T is 0.92xxxx, in this case, you may argue that selecting a threshold of 0.96 (half way between 1 and T) is likely to be safe. If T was 0.88xxxx, the 'safe' threshold could be 0.94 and so on. If all references you've seen so far are in fact legitimate duplicates, I would suggest continuing the manual procedure until you'll find a few non-duplicates.
This way to proceed is not guaranteed to be 100% accurate, it will be up to you to decide if it's good enough (and of course, you may change how you calculate the 'safe' threshold). The big limitation of 'automatic marking' is that there is no easy 'undo' route, this is why it's risky! As a way to mitigate this problem, it is possible to cancel the operation while it runs. In this way, to double check the accuracy of the process, you can lower the threshold a bit, let the system evaluate a few groups, click cancel and manually look at what happened. Note that, to save time, the engine skips already completed groups.
I am very aware that duplicate checking is complex and time consuming, so I'm hoping this message will help you a little.
Best wishes,
Sergio