SYSE.37:11 - SoTA-Echoing

For “When should reliability risk interrupt someone?”, adapt the historical 2018 SRE Alerting on SLOs multiwindow line. It improves on an immediate threshold or a long stale window by relating loss and continuing activity. Sections 4.2–4.3 rederive the numbers and retain service-specific low-traffic limits.

Compare that decision with current Cloud Monitoring burn-rate semantics, where the selected SLO, lookback, evaluation condition and notification channel are concrete implementation concerns. Reject treating either provider defaults or historical example thresholds as universal policy. The additional fit work costs maintenance but prevents a nominally correct alert from protecting the wrong interval or population.

Reopen the arrangement when task volume, objective, metric meaning, monitoring implementation or response capability changes. An old notification test does not prove today’s recipient can perform the required recovery.