The Paging Policy That Respects Sleep

Wake the on-call only when you mean it. The four rules that prevent the page-everything-just-in-case habit.

The four rules

Four rules separate paging that respects sleep from paging that destroys it. Each one is mechanical; the discipline is enforcing them.

Rule 1: customer impact only. Internal warnings go to a dashboard, not a phone; only customer-facing problems wake humans.
Rule 2: actionable tonight. Only page when the on-call can act; if it cannot be fixed tonight, the page is theatre.
Rule 3: aggregate before paging. 50 alerts on the same incident equals 1 page, not 50; the dedup is a feature.
Rule 4: runbook required. Every page links to a runbook; no runbook means the page is too vague to act on.

Without a weekly audit, the rules drift. The page log is the cheapest signal that the discipline is holding.

Weekly review. Of last week's pages, how many were real incidents that needed human action?
Threshold. Anything below 70% real-incident rate is a problem; the team is being woken up for noise.
Tighten until honest. Adjust rules until the rate clears 70%; loose paging erodes trust and burns out engineers.
Owner. One named engineer runs the audit each week; the discipline lives or dies on the cadence.

The rules are averages, not ceilings. Mistakes happen; the policy provides escape valves so noise becomes data, not just frustration.

Wrong pages happen. Some pages will be noise; the policy is the average across the quarter, not the ceiling per shift.
Document the noise. When a page fires for noise, the on-call writes it down; the note feeds next week's tuning.
Distributed accountability. Tuning is everyone's job, not just SRE's; service teams own their alert hygiene.
Quarterly retro. Aggregate the noise notes; the patterns drive structural fixes, not per-alert tweaks.