Routing all interrupts to one rotating person raises the rest of the team's uninterrupted time

Hypothesis

The SRE book's chapter on interrupts argues for "polarizing time": a person should know each day whether they are doing project work or interrupts, context switches should be assigned a cost, and tickets should not be randomly assigned across the team. The hypothesis is that adopting this structure on a small product team (four to eight engineers, no formal on-call) measurably raises uninterrupted time for the non-rotating members, and that the interrupt handler's response time stays within the team's stated target.

Prediction

Over comparable periods before and after the change: the median uninterrupted block (time between two interrupt-type events for a person, measured from calendar, chat mentions and ticket assignments) rises for non-handlers; the share of working days with at least one two-hour block rises; the time to first response on tickets and questions does not exceed the agreed target more often than before. A secondary prediction is that the handler reports fewer, not more, interruptions than the sum previously spread across the team, because batching lets similar requests be answered together.

Proposed test

1. For four weeks, log interrupt events per person without changing anything: source (ticket, chat question, request from another team), time, who handled it, time to first response.
2. Agree a response target (for example, first response within two hours during working hours) and a rotation (one handler per day or per week; the SRE book describes a week as the ideal and a day or half-day as the more practical option).
3. Route every interrupt channel to the handler: a shared alias, a chat group, ticket auto-assignment. Others redirect requests they receive personally.
4. Log the same events for another four weeks; compute the block lengths and response times per person and per role.
5. Compare distributions, not averages; report the number of events, the periods and any confounders (holidays, incidents, releases).

Status

No result claimed. Effects may depend on whether the handler has enough knowledge to answer without escalating; a handler who forwards everything reproduces the old pattern with an extra hop. Teams with fewer than four people may not have enough capacity to dedicate one person; the hypothesis does not cover them.