Designing for failure is difficult because it’s usually not practical to predict every possible thing that could go wrong. Last month we got the preliminary “spadework” out of the way. This month we begin our first feature. This feature creates a mechanism for capturing those “random and rare” failures.