This doesn't fit in to what is the general definition of a support policy/stability policy.
Perl 6 is making this a habit to take well used, defined and understood terms and give them totally different meanings. And then then expect the whole world to relearn those new meanings.
The general definition of a support/stability policy is to define what will be fixed, when it will be fixed, and till how long the fixes will be provided between two 'No going back from once made to public' releases. That's how Perl 5 defines it and all software that is released
I think what you are talking of is number of test cases failed or spec coverage affected with a particular refactoring effort. In that case a stability policy is meaningless. For a programming language, many features depend on each other.
A program has many parts, many that depend on each other. If you say you are breaking a very tiny part of it, in essence you break the whole program Eg: If you break the addition operator, you are likely to break all while loops in a program, that is likely to break nearly everything. Therefore a stability policy for any such non backwards compatible release doesn't make much sense.