It is critical to discover deployment problems and residual bugs within fraction of an hour after release. As error codes definition are stable across the releases, it is feasible to identify patterns for good and problematic situations. This way, we could develop optimized tools for automatic alarming the team when something goes wrong in the very beginning of a release. (Our current tools have data vendor-related latency and work well for monitoring and analysis in the some time in the past - last 4 hours, last week, etc. )