Glossary
Definitions as they are used on this site. Operations vocabulary is full of words that have a precise meaning in one tradition and a vague one everywhere else; where a term is genuinely used inconsistently, the entry says so rather than quietly picking a side.
Toil
Manual, repetitive, automatable work that carries no enduring value and grows in proportion to the size of the estate. Widely used to mean any work people find unpleasant, which drops the clause that makes the concept useful. Unpleasant work of fixed volume is a morale problem; work that scales with the estate is a capacity problem, and only the second one eventually sinks the function.
Interrupt load
The share of a person’s working time that arrives unscheduled and demands immediate attention. Distinct from total workload, and a better predictor of whether anything long-running ever gets finished.
Rota
The published schedule of who is covering what, when. Called a roster in some regions and simply “the schedule” in others; the three words are used interchangeably. Distinct from the rotation, which is the pattern the rota repeats.
On-call
Contested, and the ambiguity has legal consequences. In one usage it means being contactable outside working hours with an expectation of responding; in another it means an active shift worked at an unusual hour. The two differ in compensation, in working-time regulation and in how much rest they destroy. Worth stating explicitly which is meant in any policy.
Coverage
The hours during which the function is attended. Frequently conflated with capability coverage — whether the person attending can actually resolve what arrives. A rota can have complete hour coverage and no capability coverage at three in the morning.
Follow the sun
Covering continuous hours by handing work between teams in different time zones, so nobody works nights. Removes night work at the cost of a handover boundary running through everything in progress.
Handover
The transfer of state from one shift to the next. The most reliable place for operational context to be lost, and best treated as a defined payload rather than a conversation.
Escalation
Used for two different movements: upward, to someone with more authority, and sideways, to someone with more expertise. Escalation policies frequently specify the first while what is actually needed is the second.
Page
An interruption that demands immediate attention regardless of what the recipient was doing, including sleeping. The word survives the pagers. What counts as page-worthy is a design decision, not a property of the underlying event.
Page budget
An explicit limit on out-of-hours interruptions per shift that the rota is designed to absorb. Exceeding it is treated as a defect in whatever generates pages rather than as a condition the team should withstand.
Acknowledgement time
How long between an alert firing and a human confirming they have seen it. Cheap to improve and frequently reported as though it measured responsiveness, which it does not — it measures whether someone pressed a button.
Runbook
A written procedure for handling a specific, recognised situation. Distinguished here from a playbook, which is broader and describes an approach rather than steps. Many organisations use the two words in exactly the opposite sense, so the distinction is worth restating locally rather than assuming.
Deviation rate
How often somebody following a documented procedure has to depart from it because it no longer matches reality. The only measurement of runbook health that is not self-reported, and it can only be obtained by executing the procedure.
Game day
A deliberate, scheduled exercise in which a procedure or failure response is rehearsed under conditions close to real ones. The point is to discover what is wrong, so an exercise that goes smoothly has usually been designed too kindly.
Service level indicator
The measurement itself: the specific quantity being observed, such as the proportion of requests handled within a period. Often confused with the target attached to it.
Service level objective
The target value an indicator is expected to meet, chosen internally. The commitment the team manages against.
Service level agreement
The contractual version, usually with a remedy attached and usually set well below the internal objective, so that the objective can be missed without the contract being breached. Popularly used as a synonym for all three of these terms, which is how conversations about reliability become impossible to follow.
Error budget
The amount of failure a target permits over a window — the complement of the objective, expressed as time or as a count. It is not a mechanism on its own. The mechanism is the consequence agreed in advance for exhausting it.
Burn rate
The speed at which an error budget is being consumed, expressed as a multiple of the rate that would exhaust it exactly at the end of the window. A rate of one is sustainable; anything higher has an arrival date. Not to be confused with the finance sense of the same phrase.
Load shedding
Deliberately declining lower-priority work when arrivals exceed capacity, in a pattern that has been decided and published in advance. Every saturated function sheds load; the choice is only whether it does so explicitly or by letting the oldest items rot at the bottom of a queue.
Utilisation
The proportion of available capacity that is committed. Where arrivals are unpredictable, waiting time rises sharply as utilisation approaches its limit, which is why a function planned at full utilisation has no capacity for the unscheduled events it exists to handle.
Queue discipline
The rule determining what gets worked next: oldest first, highest severity first, cheapest first, loudest requester first. Every function has one. Most have never chosen it deliberately.
Blameless review
A review conducted under a specific analytical constraint: an outcome may not be explained by what a person should have done differently, because that explanation ends the investigation exactly where the useful findings begin. Frequently misread as a tone of voice, and occasionally as an absence of accountability, which is a different thing entirely.
Contributing factor
A condition without which the outcome would have been different or less severe. Preferred here to root cause, which implies a single origin that complex failures rarely have and encourages stopping at the first plausible one.
Mean time to repair
Ambiguous enough to be worth avoiding without qualification. The same abbreviation is used for time to respond, to repair, to recover and to resolve, which are four different measurements, and averaging any of them across events of wildly differing severity hides more than it shows.
Operational readiness review
A check, before something becomes the operations team’s responsibility, that it can actually be run: that it emits something useful when it fails, that a procedure exists, that somebody is accountable, and that the additional workload has been counted. The main defence against operational load arriving unfunded.
Single point of knowledge
A task or system that only one person can handle, usually undocumented and usually invisible until that person is unavailable. Detected by removing people from the schedule rather than by asking who knows what.
Fatigue
Cumulative impairment from disrupted or insufficient sleep. Relevant here because it degrades exactly the capacities operational work depends on — vigilance, judgement under uncertainty, and the willingness to escalate — and because it is a predictable output of a rota rather than an individual failing.