I think it is possible to get to sophisticated about the 'human factors' problem. It is all very well talking about isolated detail (sidestick v yoke, angle of attack meter display), but I think the real underlying issue is the pyschological response which seems to have determined that something inexplicable was happening. In a straightforward, known, simple, deterministic system which you know doesn't embody millions of man hours of design, coding and testing, there are relatively few parameters and few possible control responses. The mental approach to problem solving in such a regime is happily constrained. In a system which you know (all that training, all those sub-modes, all those interactions and dependenciess to remember) to be complex, starting to understand the problem is polluted by the possibility that the complex intervening layer is behaving in a way which you either don't understand, or which has bust and gone outside its normal limits.
It doesn't require the complex system to in fact break, or for you to forget your training, to introduce the possibility that a stressed human mind trying to start to appreciate an unanticipated possibility never achieves sufficient confident to diagnose and then act. The awareness of a complex intervening layer is in itself sufficient to frustrate intelligent problem solving, at least in a timely way.
A 'big red button' which forced direct law, with direct law behaviour a set of memory items, would seem a reasonable last resort to offer a bemused crew.
Not forgetting appropriate display of unusual trim conditions and cancellation of any warning cancellations, naturally.