The real cost of an interruption is twenty-three minutes
Attention research put a number on the cost of an interruption. Engineering calendars are built as if the number were zero.
Gloria Mark's research group at UC Irvine spent years measuring what happens when knowledge workers are interrupted, and the number that survived into common knowledge is about 23 minutes: roughly how long it takes to fully return to a task after switching away, per her published research on interrupted work. The finding has aged into a statistic people quote and calendars ignore.
Read an engineering team's week against it. A standup at 9:30, a design review at 11, a one-on-one at 1:30, a vendor demo at 3. Four sensible meetings, none longer than an hour, and between them the day contains no block long enough to reach the state where hard work happens, because each boundary charges the reentry toll. The calendar shows four busy hours. The attention ledger shows a day spent mostly in transit.
The fixes are unglamorous and work. Meeting days and maker days, or at minimum meeting mornings and protected afternoons, agreed at the team level so no individual has to defend their own wall. Interrupt budgets for the channels that page people. And a standing question in retros: which recurring meeting would we cancel if we had to cancel one. There is always an answer, and it is always the same answer from everyone, which tells you something.
The tradeoff is coordination latency: protected time means some questions wait hours instead of minutes, and the people whose job is unblocking others feel that cost personally. The reversal condition is genuinely interrupt-driven work, operations during an incident, support rotations, where the interruption is the job. For everyone else, the 23 minutes are being spent. The only question is whether anyone is counting.