It's just a rule of thumb, like the "rule of 7" when it comes to cognitive load. The point is to continue past the surface issue until you find the real problem

It's just a rule of thumb, like the "rule of 7" when it comes to cognitive load. The point is to continue past the surface issue until you find the real problem. Fixing a lower-level problem (like not having automated testing) has much more value than fixing a higher-level problem (like fixing regressions when they occur in your software).