An Implementation Model for Rational Protocol Conformance Behavior

So... What is the current consensus? Fixing this by the 6.0 release? :relaxed:
Because I have just come to a great idea, which is: let's add to a process of archetype construction a notion of context (that is the current module, aka the most recent refinement).

//module S
protocol A { var uo: Int {get} }
extension A { var uo: Int {42} }
struct Alpha: A {} 
//this has archetype S.Aplha with witness from S.A

//module T
import S
extension Alpha: A { var uo: Int {69}}
//this has archetype T.Alpha with witness from T.Alpha
//Yes, single redeclaration in different module should be allowed

From here, we dance: compiler builds the internal code from external modules and then it becomes a different realm (nothing can touch it from public domain), then it selects the witnesses (redeclarations) in the current root (in the example above it is module T), and after, it descends the imports graph, picking the witnesses of most recent refinement from external modules for missing conformances.
What do you think about that kind of a monomorphizing solution?

Checkout this btw