On-call is a budget, and yours is being overspent

Teams do not get slower. Codebases get older and rotations get thinner, and of the two, the rotation thins more quietly. Someone leaves, the roster drops from six to five, then a hiring freeze, then four. No meeting decided that each remaining engineer would carry fifty percent more nights. The math decided it, and math does not send calendar invites.

The spending shows up in disguised forms first. Sprint work from whoever is on call this week quietly stalls, so velocity wobbles without pattern. The most experienced engineers, who anchor the rotation, drift toward burnout or toward teams without pagers. Deploys migrate away from afternoons because nobody wants to own the evening, which lengthens lead time, which someone then diagnoses as a delivery problem. We have been called in for the delivery problem and found the pager problem more times than we can politely count.

Reading the budget takes an afternoon: pages per shift, pages that woke someone, time to hand off, roster depth against the coverage math, and the one number nobody tracks, how many of the pages required no action. Actionless pages are pure spend, attention withdrawn at 3 a.m. and nothing purchased, and in most systems we read they are the majority. Deleting an alert that has paged nobody into useful action for a quarter is the cheapest morale work in engineering.

The tradeoff in treating on-call as a budget is that budgets force refusals: a roster of four may simply not cover two services, and saying so means someone above the team accepts either risk or cost. That conversation is the point. The reversal condition is a service quiet enough that the pager is ceremonial, and if so, the kindest audit finding is to say that out loud and stand the rotation down.