How do we learn from failures in an open source way?