Hello JB (and Phil !),
I am experimenting with your (great) tool since few months now, and I find it very pleasant to use : not only simple/ergonomic, but also quite powerfull... many thanks to you and to Phil for sharing it (and spending your time) with us !

I was reading the answer you made to Rennie, and I myself decided to use the 1st approach you described when time comes to add some legend to my views, as I pay particular attention to keep my repository "clean" (I don't like to "pollute" it with "fake", "theoretical" or "hypothetical" concetps). So I am using "visual" notes, groups and connections as much as possible...
That said, I was recently wondering why it is actually not possible to "mimic" composition/aggragation/assignment relationships through visual connections, and also why it is not possible to link together two archimate concepts (application components, interfaces, ...) using these "visual connections" (which are only visible in views, without creating underlying object in the model repository).
Indeed, I have a particular use case in mind when asking that, which go beyond the topic of legends and which must be frequent in most organizations : when I'm working on a new project (at solution architecture level), I am often (no... always !

) asked to identify, design and evaluate all possible architecture scenarios (most of times, there are 2 or 3 candidates for the "target architecture").
If "visual only" connections were possible between archimate concepts (and composition/aggragation/assignment arrow styles allowed, to mimic all kind of relationships), then it would be possible to model all candidate architecture scenarios for a particular project directly on the "reference model" of the organization, in one or several views (doesn't matter), without polluting it with some sort of hypothetical data flows / triggering relationships (/ whatever relationship between concepts) which will never exist in the future (because only 1 of the 2 or 3 candidates will be implemented at the end).
As it is not possible (for now ?

), the only work-around I found was to use "official" archimate relationships between concepts... and then to maintain two archi models in parellel : one "reference model", and another one dedicated to projects... Which is not optimal (to my point

) because you have to play and juggle all times with the import/export functionality (which can be a little bit tricky when you are in a team with several architects using co-archi, a "reference model" that evolves a lot (being done "from scratch"), and literally tons of projects to handle in parallel !

.
What's your thoughts about this ? Is my idea totally weird ? (maybe there are some potential repercussions that I don't see...) ?
Is the work-around I use the only way to go (or have you got some advices to make it easier to handle) ?
Thank you in advance for your answer and.... keep up the (very) good work !

Regards,
Eric.