ChessMoves AI

The three things you should never have to go looking for — a stale/failed backup, an unreviewed security queue, or a spike that looks like abuse or exfiltration. Fuller telemetry lives on the Metrics page.

Nightly PostgreSQL dump → local + offsite Storage Box. A backup older than the threshold, or a recorded failure, is the alarm.

Unreviewed queues that need a human. Non-zero "pending" is the signal.

Derived from login events, admin actions and signups. Brute-force = failed-login spikes across many IPs; account takeover / sharing = one account signing in from many IPs; exfiltration = unusual admin-action or signup volume.

The panels above answer “is someone attacking us right now”. This answers “is the hardening actually in place” — derived from the running config and the source on disk, never from a hand-kept checklist, because a typed checklist is how you end up asserting a limiter that was removed a year ago.

One box runs everything, which is a single point of failure the moment an uptime figure is promised to anyone. The useful question is not “should we scale out” but “what breaks the second a second replica starts” — so this lists that, separating what is measured against the running system from what is reviewed by reading the code, because “this map is never replicated” is not something a process can prove about itself.