Personally speaking, isn't that exactly the value of having a type system in the first place? If it still type checks after the refactor it ideally* is still working. Unless of course some system boundaries changed or there is some dynamic component.
I suppose we both come from different philosophies here. I write out the types for a program first, then the behavior follows through. If I cannot properly determine the types I escape with dynamism.
You seem to dynamically write the system up and then determine the types, do I interpret that right?
The current setup might be a better experience then, but I think the default matters here.
In my little bubble I've encountered more libraries with broken typing since type checking has been reduced, so I perceive it as a net-negative. You end up having to explain that you need to take manual precautions to actually get type checking.
Of course this could just be due to the growing user-base and the higher probability of hitting a "wrongly typed" library, so it will remain to be seen how much impact it has.
I'm used to the --check flag by now :)
* How ideal depends on how expressive the type system is.