This flag seems like a bad idea. Can someone explain what measure will determine that results are valid after violating the desired tolerance for the governing equations? How critical is the violation? It seems that there can be repeated violations? Can every timestep in a simulation violate and complete? What if the residual is very high? Is there a criteria for termination then?

What was the previous behavior if this option was enabled? How does this change interact with other checks within the physics solvers themselves, e.g.

If we keep this, I would suggest two changes:

1. Modify the XML key so the user is very clear they are entering a danger zone. I would suggest making the string something like DANGER_ZONE_allowNonConverged so the user has to markup their input deck with an obvious "I know what I am doing" flag.
2. We could force modify the output file string to automatically append something like myOutputName_UNRELIABLE so that GEOS acknowledges it itself doesn't trust the results.

I can see selected situations where for debugging purposes a user may want to keep the simulation running to see what happens and gather more info, but it has to be very clear that they are taking ownership of any poor quality results.