Avoiding false reassurance
An empty alert list can mean all rules are normal, a source is unreachable, rules are absent, or data is missing. The application must expose source failures and never infer universal health from that one list. Likewise, transport uptime and service correctness are independent signals.
Threshold duration
Sustained windows filter scheduler noise and short bursts. They should not delay fast availability alerts: a required service outage needs a shorter window than high CPU. Backup deadlines depend on schedule rather than a universal fixed duration.
Independent monitoring
The portal cannot reliably report its own total outage. Its external synthetic check and cloud alert evaluation form the independent failure detector. Notification routing remains in the existing provider rather than adding another competing alert engine.
Complete rule catalog
The public Alert rules page lists configured rules, including healthy and paused rules. It reads actual evaluation intervals from Cloud rule-group metadata; observed groups evaluate every 60 or 300 seconds. Pending duration and lookback are separate fields. Silenced rules remain enabled unless paused. Private expressions require owner diagnostics. Missing evaluation metadata is shown as unknown rather than inferred from query resolution.