As site scale and traffic change, once-good thresholds decay: too many alerts go unread, too few miss real problems. Thresholds may be adjusted by definition, never casually just to make alerts stop.
Periodically re-check the baseline against status-code distributions: the floor of 5xx share, timeout ratio and crawl-error volume moves with scale, so recalculate thresholds on the same definition. Keep before/after records and reasons; security thresholds can reference Web Security Testing Guide items without treating scan noise as a real alertable event.
Alerting is a recurring action: if numbers don't move, record them as-is — no direction change, no extra spend. This article issues no verdict on any single threshold adjustment.