A sysadmin who appears constantly busy, who fights incidents on repeat, who responds to alerts at all hours, gives an impression of dedication and high workload. This impression is misleading. A well-designed system does not produce serial incidents: it anticipates failures before they become visible. Visible activity is therefore not proof of good work. Most often, it is the symptom of an earlier failure to anticipate.

Site Reliability Engineering, as documented by the teams who formalised the discipline at Google, names this phenomenon precisely under the term "toil": manual, repetitive operational work that grows proportionally with system scale and produces no durable value once executed. The SRE discipline sets an explicit limit on this work — it must not exceed a reasonable share of total time, the rest being devoted to engineering that reduces this repetitive work in the future.

It is midnight on a Black Friday. A company's e-commerce traffic has multiplied eightfold within hours. In the virtual war room, part of the technical team cycles through manual restarts, watches dashboards blinking red, and negotiates live with the cloud provider's support to secure more capacity. The SRE responsible for the payment infrastructure is not in that room. They are asleep.

They are not asleep out of negligence. They are asleep because the previous six months were dedicated to exactly this moment: load tests simulating ten times normal traffic, automated failover drills repeated in pre-production, an auto-scaling policy validated against real scenarios rather than theoretical assumptions. The Black Friday spike is not an unforeseen event for their system — it is a scenario already rehearsed several times, under conditions harsher than reality.

This scenario, deliberately composed to illustrate the principle rather than to recount one specific real case, reflects a documented and widely taught practice in reliability engineering: regular failure drills, conducted upstream and outside periods of criticality, replace improvised crisis management with already-rehearsed execution. Silence in production is not a mystery. It is the measurable result of work correctly sequenced over time.

Let's say out loud what many have practised quietly since remote work became the norm. The video call where one wears a shirt, sometimes a tie, above the belt, and simple underwear or shorts below the camera frame, has become a universal scene of remote work — funny because it is true, and true for a reason that goes beyond mere convenience.

Sleep and cognition science have documented this mechanism for decades: a short nap improves alertness, working memory and problem-solving ability in the hours that follow, unlike forced sustained attention, which progressively degrades the quality of judgment. An SRE who takes twenty minutes of sleep in the afternoon, or a clean break between two tasks, does not work less over the day. They avoid the degraded mode of operation that produces configuration errors, rushed on-call decisions, and ultimately new incidents.

The activity report, in its most common form, asks people to justify time spent rather than results produced. This is, first and foremost, a metric error. Labour sociology documents this phenomenon under the name presenteeism: valuing visibly occupied time over actually produced results, which pushes individuals to favour the appearance of activity over effectiveness.

This first imbalance, once embedded in a team's culture, does not remain a mere administrative curiosity. It produces a second, deeper effect: a competitive climate around appearing busy. If the recognised metric rewards visible activity, team members are structurally incentivised to display themselves as more loaded than their neighbour, rather than to cooperate so that the shared infrastructure has as few incidents as possible.

The timesheet is therefore not merely a symptom of an initial bad metric. Once this climate is established, it becomes a structural problem in its own right, one that keeps producing its effects even if one attempts to fix its wording without changing the culture that gave rise to it.

An SRE does not owe their organisation 35 hours of visible presence per week. They owe the optimal availability of the service they are responsible for. These two obligations rarely overlap in practice, and this document has tried to show why: a well-designed system keeps its guardian silent, while a poorly designed system keeps its guardian visible, busy, and apparently indispensable at every moment — for the wrong reasons.

The practical consequence of this analysis is not the abandonment of all SRE performance measurement. It is a shift in criteria: measuring the service's actual availability, the reduction of repetitive work over time, and the quality of anticipation — rather than the volume of visibly occupied hours. An SRE who appears to do nothing because their infrastructure works for them is not an employee falling short. They are the most tangible proof that their earlier work was done correctly.