If you want to add support for a new kind of register class in GCC, do you refactor everything as you make your change? No, you'll break a dozen architectures you've never heard of. Instead, you add some logic in a self-contained way, conseratively, assuming you don't fully understand the corner cases of the existing implementation. It's much easier to understand new logic as a delta of existing behavior, which may be then partially specified, than to reimagine the logic from first principles.

That's why the law is full of "But in cases under subsection C where Z applies..." You don't refactor subsection C because you may break it. Better to specify the additional behavior in a self-contained way. You can summarize the behavior of a legal code in diagrams, the same way you can diagram the high level operation of GCC's register allocator. But the diagram isn't the code, and the diagram can't be canonical.