We rebuilt our provisioning queue

Creating a hosting account touches five systems: the control panel, DNS, SSL, mail and billing. When any one of them fails, the customer sees "pending" and support sees nothing. Every order now creates a job with an explicit step ladder — assign server, create account, configure DNS, issue SSL, create mailboxes, send credentials. Each step records its own status and duration. Two consequences matter. Failures are specific: "CAA record blocks Let's Encrypt" instead of "provisioning failed". And retries are safe, because a step that already succeeded is skipped rather than repeated. Structural failures — a full server, a registry rejection — stop retrying immediately and raise a person, rather than burning four attempts over eight minutes on something no retry can fix.