---
id: ops.operate-and-support
title: Operate and support
kind: playbook
domain: operations
description: Run a system in Support - take it over after the readiness review, keep a calendar of routine duties, take in work by class, handle incidents, review service levels every month and retire the system when the owner decides, without ever acting on a live system without a backup proved by restore.
version: 0.1.1
status: review
authority: uask
applies_to: [core]
sources:
  - title: 'GOV.UK Service Manual, "How the live phase works" (supporting the service in a sustainable way and continuing to improve it)'
    url: https://www.gov.uk/service-manual/agile-delivery/how-the-live-phase-works
    license: OGL-UK-3.0
    accessed: 2026-11-21
  - title: 'GOV.UK Service Manual, "Retiring your service" (tell users, deal with the data)'
    url: https://www.gov.uk/service-manual/agile-delivery/retiring-your-service
    license: OGL-UK-3.0
    accessed: 2026-11-21
  - title: 'Site Reliability Engineering, chapter 5 "Eliminating Toil" (manual, repetitive work is tracked and reduced)'
    url: https://sre.google/sre-book/eliminating-toil/
    license: CC-BY-NC-ND-4.0
    accessed: 2026-11-21
requires: [process.project-lifecycle, mgmt.roles-and-decision-rights, ops.production-readiness-and-handover, ops.service-levels, ops.incidents-and-postmortems, ops.backups]
related: [process.deliver-a-release, quality.quality-release, mgmt.delivery-metrics, arch.risks-and-technical-debt, mgmt.governance-tiers]
last_reviewed: 2026-11-21
review_every: 6
---

## Summary

A system in Support is a system with users, data and someone who answers for it. This playbook says how that someone works: takes the system over after the readiness review, keeps a calendar of the duties that repeat, takes in
every piece of work by class, handles incidents and learns from them, looks at the service levels every month, and, when the owner decides, retires the system without losing what must be kept. It does not repeat the rules of the items
it orders; it gives them an order and a rhythm.

## Rules

- **R1.** Support MUST NOT start until the system has passed the production readiness review, or the owner has accepted the open lines in writing (`ops.production-readiness-and-handover#R12`, `ops.production-readiness-and-handover#R13`,
  `process.project-lifecycle#R9`).
- **R2.** The owner MUST name who supports the system and who stands in for them; the lead MUST record both with the date. A system without a named supporter is not in Support.
- **R3.** The supporter MUST keep a calendar of routine duties, each with an owner and a date: the restore test of the backups (`ops.backups#R6`), the renewal of certificates, domains and licences, the review of dependencies against known
  exploited vulnerabilities (`quality.quality-release#R16`), the access review, the review of the runbook and the monthly service-level review. A duty that nobody has done twice in a row SHOULD be automated or dropped.
- **R4.** All work for the system MUST arrive in one place, and each piece MUST be given a class and the class recorded with it: fix, change or system change (`process.project-lifecycle#R10`). When in doubt the higher class MUST be taken.
- **R5.** Before the first change to a live system, a backup proved by restore MUST exist, and changes MUST be built and verified on a staging copy, except for a change the lifecycle rules name as needing none
  (`process.project-lifecycle#R11`, `process.project-lifecycle#R12`). The window for a change that affects users SHOULD be chosen from traffic data, with someone able to roll back until verification ends (`quality.quality-release#R18`).
- **R6.** A fix MUST go through Build and Release with a failing check first and a regression test (`process.project-lifecycle#R10`), and MUST pass the same gate and the same backup as any release; only the steps that the release checklist marks as shortened MAY be shortened, and a hotfix, a defect that cannot wait, follows `quality.quality-release#R22`. A change MUST run as a release
  (`process.deliver-a-release`). A system change MUST go back to the Blueprint.
- **R7.** Incidents MUST be handled as `ops.incidents-and-postmortems` says, and the service levels MUST be reviewed with their numbers each month at the standard tier and above (`ops.service-levels#R10`).
- **R8.** A known-exploited vulnerability in a dependency of a live system MUST be fixed or the risk MUST be accepted as a recorded exception by a named person, within a time the project states in its runbook
  (`quality.quality-release#R16`).
- **R9.** The review of the month MUST be one page: the service levels against their targets, the incidents and the state of their actions, the routine duties done and missed, the open risks and debts
  (`arch.risks-and-technical-debt`), the work taken in by class and the time spent, and the decisions the owner is asked for. It MUST end with at most one change to the way of working, or "no change".
- **R10.** The time spent on the system MUST be logged by class so that the owner sees what the system costs to keep. Work that repeats SHOULD be turned into a runbook step or a script.
- **R11.** An agent MAY take work of the fix class under a standing rule of the owner that names the kinds of work and their limits (`mgmt.roles-and-decision-rights#R17`). An action that `mgmt.roles-and-decision-rights#R9` reserves to the owner
  (production of a live or client system, money, domain names and DNS, keys and access, deletion of data) MUST NOT be taken without the owner's yes or such a rule.
- **R12.** For a client's system the client's reports MUST reach the intake through the owner or through the channel the owner has set, and the client MUST NOT instruct agents (`mgmt.roles-and-decision-rights#R7`).
- **R13.** Retiring a system MUST be the owner's decision. The plan MUST say how users are told, which data is exported, kept or deleted and for how long (UK GDPR retention), where addresses are redirected, which access and accounts are removed, which
  backups are kept and for how long, and who confirms each. Data MUST NOT be deleted before the last backup is proved to restore and the owner has said yes.
- **R14.** The phase MUST open with a recall step and close with a record step (`process.project-lifecycle#R13`): at take-over the earlier work on the system is read, and at retirement or hand-back what was learnt is recorded with where each lesson goes.

## Stages

| # | Stage | Artefact | Who decides |
|---|---|---|---|
| 1 | Take over | hand-over record, named supporter, calendar | lead passes the review; owner places the support |
| 2 | Run | calendar done, logs, backups tested | supporter |
| 3 | Take in work | classed work, plans of fixes, release specs for changes | lead; owner for scope and money |
| 4 | Respond | incident log, postmortem | incident lead; owner for owner-level actions |
| 5 | Review monthly | one-page review | lead; owner for the decisions on it |
| 6 | Retire | retirement plan and record | owner |

### 1. Take over

- **Entry.** The system has passed, or is about to pass, the production readiness review.
- **Do.** Start with the recall step (R14): the earlier work on the system, its release records and its open findings. Pass the review (`ops.production-readiness-and-handover`). Ask the owner to name the supporter and the stand-in (R2).
  Build the calendar of R3 from the hand-over record: the dates of renewals, the next restore test, the next review. Add the first month's review to it.
- **Outputs.** The hand-over record, the named supporter and stand-in, the calendar.
- **Exit.** R1 to R3 are met and the owner has recorded who supports the system.

### 2. Run

- **Entry.** The system is in Support.
- **Do.** Do what the calendar says on its dates and record it. Watch the monitoring and the backup alerts. Read the logs that the runbook names. Keep the runbook true: whatever a supporter had to ask for or guess is added to it within a week
  (`ops.production-readiness-and-handover#R15`). Track the time that routine duties take and look for what can be automated (R10).
- **Outputs.** The calendar's duties done and recorded; an up-to-date runbook.
- **Exit.** Each duty of the period is done or the miss is recorded with a reason.

### 3. Take in work

- **Entry.** Work arrives: a request, a defect, a change the owner wants, a finding.
- **Do.** Put it in the intake and give it a class (R4). Fix: write the plan of the fix, the failing check, the backup and the rollback, build it, release it through the same gate and backup, shortening only what the release checklist marks as shortened. Change: write a release spec and run `process.deliver-a-release`. System change: go back to
  the Blueprint with the owner. Re-class if the work grows. Ask the owner about scope or money, with a default (`mgmt.roles-and-decision-rights#R12`).
- **Outputs.** Classed work, the plans or specs the class needs, and the log of the time spent (R10).
- **Exit.** Each piece of work is done and released, or declined, or moved to a higher class, with the class recorded.

### 4. Respond

- **Entry.** An incident is declared (`ops.incidents-and-postmortems#R2`).
- **Do.** Follow `ops.incidents-and-postmortems`: one incident lead, restore first, a log, the owner told in time, the client told through the owner. After service is back, write the postmortem where its criteria are met and turn each
  action into work of the right class.
- **Outputs.** The incident log and the postmortem; actions in the intake.
- **Exit.** Service is restored, the owner is told, and the postmortem and its actions exist where required.

### 5. Review monthly

- **Entry.** The date of the review in the calendar.
- **Do.** Write the one page of R9 from the records: the service levels against their targets, the incidents, the duties, the registers, the work and the time. Decide at most one change. Send the decisions the owner is asked for with options and a
  default. Compute the delivery metrics that apply (`mgmt.delivery-metrics`).
- **Outputs.** The one-page review.
- **Exit.** The review is filed, the owner has what they must decide, and the next review is in the calendar.

### 6. Retire

- **Entry.** The owner has decided to retire the system, or to hand it back or to another supporter.
- **Do.** Write the retirement plan of R13 and have the owner approve it. Tell users as the plan says. Export or keep the data as it says. Redirect the addresses. Prove the last backup restores before any deletion. Remove access and accounts. Keep the
  backups for the agreed time. Close the calendar. Write the record step: what the system taught, and where each lesson goes.
- **Outputs.** The retirement plan, its record, the list of what was kept, and the lessons.
- **Exit.** The plan is carried out and confirmed item by item, and the system no longer appears in the calendar.

## Guidance

### When to use

For every system after its first release that has passed or is passing hand-over, from a one-page script that a client depends on to a shop with orders. For a system that nobody supports, use stage 1 to decide that fact explicitly
(R2); do not call it supported.

### Tailoring by tier

At the lightweight tier the calendar is a short list, the monthly review a few lines and the postmortem a paragraph; the rules about backups, owner actions and classes stay. At the standard tier everything applies. At the high-assurance tier add a rehearsed rollback
for each release, a second reviewer for changes (`mgmt.governance-tiers#R8`) and keep the evidence of each review. The owner's decisions stay at every tier.

### A calendar that works

Put each duty on a date and a name: "restore test, first Monday of the quarter, the supporter"; "domain renewal, 1 March, the owner"; "access review, each quarter, the lead". A duty with no date is a wish. At the review of the month look at what was missed, not at what was done.

### The first month

The first month after hand-over is where the runbook is tested by real use. Expect to add to it each time someone has to ask for something, and use the first review to look at how many such questions there were.

### Common mistakes

- Starting Support with open lines of the hand-over and no dates for them.
- Treating a defect fix as a favour outside the process; the class and the log are what make the next estimate possible.
- A restore test that has not been done since hand-over.
- Deleting data on retirement before the last backup is proved to restore.
- An agent that takes a production action because it worked last time; a standing rule must name the action.

## Explanation

**Why a rhythm.** Support is where a system spends most of its life, and the work that matters most there repeats. GOV.UK describes the live phase as supporting the service in a sustainable way and continuing to improve it. A calendar and a monthly page
keep both in view: the duties that stop the system rotting and the improvements that keep it useful. **Why track routine work.** The SRE book defines toil as manual, repetitive, automatable work that scales with the service; counting it is how a team sees what to automate before it
fills the week. **Why retirement is a stage.** Services that are switched off without a plan leave users with dead links and data nobody can account for; GOV.UK's retirement guidance asks for users to be told and for the data to be dealt with under the data protection
policy. A retired system still has obligations for as long as its data exists.

## Examples

**Good.** A small UK shop is handed to a support team on 1 December with one waived line (a certificate renewal date). The owner records the supporter and stand-in; the calendar holds the restore test on the first Monday of each quarter, the certificate on 14 February, the domain on 1 March and the monthly review. In
January a request to "show delivery dates in emails" is classed as a change and runs through `process.deliver-a-release`; a typo in the returns page is a fix with a plan of one line and a log entry. In February an outside check of the order page fails for 40 minutes; the supporter
declares an incident, rolls back the last release, the owner is told in 15 minutes and the postmortem adds a scripted test payment to monitoring. The one-page review shows 99.7% against a target of 99.5%, 11 hours of support time by class, and one change to the way of working.

**Bad.** The agent that built the site says "it is live, ask me if anything breaks". There is no named supporter, no calendar, and the certificate lapses on a Sunday. A client's request for "a small change" to the checkout is taken as a fix and edits a table on the live database with no backup proved to restore.
The owner learns of it from the client, and nothing is recorded that would tell the next supporter what happened.

## Sources

- [GOV.UK Service Manual, How the live phase works](https://www.gov.uk/service-manual/agile-delivery/how-the-live-phase-works) — Open Government Licence v3.0; supporting the service sustainably and continuing to improve it. Accessed 2026-11-21.
- [GOV.UK Service Manual, Retiring your service](https://www.gov.uk/service-manual/agile-delivery/retiring-your-service) — Open Government Licence v3.0; telling users and dealing with data when a service is retired. Accessed 2026-11-21.
- [Site Reliability Engineering, chapter 5, Eliminating Toil](https://sre.google/sre-book/eliminating-toil/) — Google, CC BY-NC-ND 4.0; linked and summarised only, not copied: toil as manual, repetitive, automatable work. Accessed 2026-11-21.
- The delivery rules of the owner's support practice (a backup proved by restore, staging, a rollback written before the deploy, a closing record), described here as a pattern without client data.
