Sounds like you've been busy! I envy your activity, skills, and the time you have been able to put toward this topic.
As for a text only modeling tool, it will be interesting to see how how completely you can model various types of domains (and to what modeling objectives), using only text based methods. My guess is that you may be able to cover all the ground serviced by a text and graphics based tool, but there will be instances where a graphical representation would be much more efficient and intuitive (just as there are times were text is better). I think I mentioned before, that a system that would let the modeler glide between text and graphic representations, on the fly, would be optimal. In order to get there, a complete representation in each form would be required. Your effort should go a long way toward showing how feasible that objective really is.
It's likely going to come down to personal preference, but I for one, would like to have fuller text based support for fact entry. Good luck with your efforts. Hope we'll have a chance to see for ourselves how usable a text only fact based modeling tool can be.
My opinion on reverse engineering relational models to conceptual models is that it's not feasible for anything but the most simple and obvious schema. As I think you pointed out, something is always lost in the process. A conceptual model is formed by intention. A data structure is only the result of some intension (deliberate, well thought out, or otherwise). Starting with the result, reverse engineering can only propose one set of intentions that would (at best), have the logical conclusion resulting in that data structure - how can the process determine if that set is the correct one? The marketing incentive is obvious: leverage the data structures you have by putting them on a rational and verifiable basis - retroactively! I just don't think the information flow map can be made to go in that direction. When the Russians reverse engineered interned U.S. B29 bombers during WWII, the knew the thing was intended to fly, carry bombs and drop them on targets. Only a full understanding of the machine's design intentions made engineering working copies feasible. BTW, I'm guessing you've tried to reverse engineer ORM diagrams that you modeled and later generated using various ORM tools. Ever seen the R.E. result look anything like the conceptual model you started with? Also, you can only know if the R.E. model is right if you fully understand the U. of D. - and if you do understand it, you can surely forward engineer (model), it anyway. Well, that's my take (rant), on the value of reverse engineering to conceptual models.