Railway's post mortem describes things that would have never made it past a whiteboarding session at AWS (multi tenant builds, overprivileged credentials, and a lack of separation between systems).

The post goes through each one. Railway's report was detailed enough to make this possible, and they deserve credit for owning up to the incident and impact. But hopefully this is a moment of reflection as well, because their report also exposes some downright silly engineering decisions that we were pretty surprised and disappointed to unpack.