Why do experienced teams track on-call interrupt hours rather than page counts alone, and what does a single overnight page actually cost the next day?

## The problem with counting pages Pager load is usually first measured as pages per shift, and that is the right starting point. But a page count is a count of discrete events, and it silently assumes every event is equivalent. Real shifts violate that assumption badly: - A self-clearing alert acknowledged from a phone in ninety seconds. - A three-hour incident with a war room, a mitigation, and a write-up. - A page at 03:10 that lasted twenty minutes but ended the night's sleep. Counting these as "three pages" produces a number that is technically true and operationally misleading. It is also gameable in the wrong direction: a team can look busy while doing nothing, or look quiet while being destroyed. ## What interrupt hours capture Interrupt hours measure the wall-clock time on-call actually consumed. The components worth including: - **Response time** — page to stand-down, including investigation that led nowhere. - **Follow-up** — the write-up, the ticket, the handoff note. Real work, often invisible. - **Re-focus cost** — the period after an interruption before deep work resumes. This is not folklore; frequently cited research on interrupted work measures task resumption in the region of twenty-plus minutes, and engineering work is close to the worst case for it. The consequence is arithmetic that surprises people. Five "cheap" daytime pages, each five minutes of actual handling, is not twenty-five minutes gone; with re-focus it can consume the productive part of a day. That is why a rotation can report a comfortable page count and still deliver nothing. ## The overnight multiplier A page that wakes someone has a cost profile of its own. You pay the incident itself, then a next day at reduced capability, and then a compounding effect if it happens on consecutive nights. There is no precise universal multiplier and you should not invent one in an interview; what you should say is that overnight pages are tracked and budgeted *separately* from daytime pages, and that the norm which follows is comp time — a late start or a day off after a bad night — treated as automatic rather than as something an engineer has to ask a manager for. The second-order effect matters more than the first: a team that pays for overnight pages properly has a real incentive to stop generating them. A team that treats them as free is quietly financing its noisy pager with unpaid human cost, which shows up later as attrition and as people declining to join the rotation. ## The decision this metric drives All of this exists to settle one recurring argument: should this condition page a human now, or become a ticket for the morning? The instinct of the person who wrote the alert is almost always "page me, I want to know". Interrupt cost is the counter-argument, and it is quantitative: this alert fires about four times a week, roughly half of them overnight, and it has never required action before business hours. That framing converts a preference contest into a cost comparison. The rule that falls out is simple: **page only when a human must act within minutes to prevent or limit user-visible harm.** Everything else — capacity trends, degraded redundancy that still has headroom, batch failures with a retry window — is a ticket. The cost model is what gives you the standing to enforce it. ## Where this fits the wider budget Interrupt hours also connect on-call to team capacity planning. If the on-call engineer's interrupt hours routinely exceed what one shift can absorb, the overflow does not vanish — it spills onto teammates, who are then also interrupted, and the team's project throughput collapses in a way that looks mysterious on a burndown chart. Google's SRE model handles this by capping the proportion of time spent on operational work outright, so that engineering capacity is protected by policy rather than by hope. Whatever cap you choose, the value of measuring interrupt hours is that you can tell when you have blown through it. ## Anti-patterns Counting only pages that were acknowledged misses the ones that auto-resolved but still woke someone. Excluding the write-up understates the real cost by a large margin. And averaging interrupt hours across the whole team rather than per shift returns to the original sin: hiding one person's terrible week inside everyone else's quiet one.