Incident Burnout Watch

Long incident weeks burn out engineers. The signs to watch and the protocol to intervene.

Signs

Burnout signs are visible if someone is watching for them. Sleep loss, mistake patterns, and social withdrawal are the recurring triad. By the time the engineer notices it themselves, the team has usually noticed weeks earlier.

• Three or more sleepless nights in a week. Visible sleep deficit. Strong indicator of unsustainable on-call load.
• Mistakes a rested engineer would not make. Tired-error pattern. Often the first thing teammates notice in code review.
• Withdrawal from team interactions. Stops attending standup, retreats from review threads, social pullback.
• Named watcher per team. Manager or peer explicitly responsible for noticing. Without ownership, signs go unaddressed.

Intervene

Intervention is the manager’s job, not the engineer’s. Pull them off rotation, activate the backup, make rest mandatory. Engineers self-regulate poorly under sustained on-call load.

• Pull off-rotation immediately. Do not wait for “they’ll be fine.” The cost of acting early is small; the cost of waiting is high.
• Activate the backup. Rotation coverage continues; the engineer rests. Backup-on-call exists for exactly this.
• Mandatory rest. Manager-enforced break. Engineers will keep responding to alerts unless told to stop.
• Documented burnout-protocol runbook. Per-team the protocol so manager decisions stay consistent across rotations.

Recover

Recovery is staged. Two days off is the floor; some engineers need a week. Return is phased through shadowing so the engineer does not re-enter the burnout state on day one.

• Two days off minimum. Floor, not ceiling. Some engineers need a full week before returning.
• Phased return. Shadow-first, then back-to-rotation. Avoids re-triggering the same load that caused burnout.
• Followup check-in. Manager’s post-return conversation catches incomplete recovery before it cycles.
• Burnout-trigger review. Per-recovery the “what caused this” review. Systemic causes get fixed; one-off recovery does not.