The monitoring that watches the Authgnosis automations was running, and it was reporting healthy, and it was still missing something. It watched whether the scheduled jobs had run. The email the website sends when someone fills in the contact form is not a scheduled job, so it was never in the picture at all. Four checks have now been added to close that gap and three others like it, and one further gap has been written down as deliberately not covered rather than half covered.

• The website's email is now proven by outcome, not by asking it whether it is well. A form submission leaves two independent traces: a record in the CRM, created by one system, and a notification email, sent by a completely different one. The check reads both and pairs them up. A record with no matching notification means the email side is failing, which is exactly the case the earlier checks had no way to see, and it now surfaces within a day. That independence is the whole point and it is easy to lose by accident, because the CRM side also sends its own notification. Counting that one would let the system vouch for itself, so only the website's own email is accepted as evidence.
• A quiet week stays quiet. If nobody submitted the form, there is nothing to pair up and the check says nothing. That restraint is deliberate. The earlier generation of this monitoring asked whether a file had been touched recently, which answers a useful question for something that runs on a timetable and a meaningless one for something that only happens when a person acts. Inventing a signal where none should exist is how monitoring teaches people to ignore it.
• New automations can no longer be born unwatched. The list of jobs being monitored was maintained by hand, and in August three jobs that were running were found to be missing from it, which is why a healthy summary kept printing while one of them was dead. The list was corrected then, but nothing prevented it drifting again. Now every running job is checked against the list automatically, so anything added in future is either covered or reported on the next run.
• The records that authorise email to leave the domain are checked for drift. A handful of DNS entries decide whether mail sent as Authgnosis is accepted as genuine. If one is edited, expires or is overwritten, outbound mail starts failing silently and the first sign is a client who never replies. The expected values are now written down in a reviewable file and compared on every run. The check reports drift and deliberately cannot repair it: a wrong automated edit to those records would break every outbound message for hours, which is a far worse outcome than a report nobody has read yet.