Production readiness and hand-over
- Status
- Review
- Version
- 0.1.3
- Authority
- UASK standard
- Applies to
- Core
Summary
A system is handed over to Support when it can be run, restored and changed by someone who did not build it. This checklist is the production readiness review of
process.project-lifecycle#R9: nine lines, each testable, passed or waived with a reason, owner and date, and recorded in one hand-over record. The lead decides
whether the system passes; the owner decides who supports it. A system that holds data nobody can recreate does not pass without a restore that has been tried, unless the owner accepts the risk in writing.
Rules
Before the review
- R1. The owner MUST decide, and record with a date, who supports the system after hand-over: the team that built it or a separate support team
(
process.project-lifecycle#R9). The review MUST NOT start before that is known. - R2. The review MUST be done by someone other than the builder of the system at the standard tier and above, and by two reviewers at the high-assurance tier, by analogy with the second reviewer of a release
(
mgmt.governance-tiers#R7,mgmt.governance-tiers#R8). At the lightweight tier the builder MAY do it and MUST say so in the record.
The lines
- R3. Runbook. A runbook MUST be in the project’s repository or in a place its supporters can reach without the builder. It MUST say how to start, stop and deploy the system, how to roll the last release back, where the logs and the data are, what the usual failures look like and how to act on each, and whom to call.
- R4. Backups. The system MUST meet
ops.backups#R1toops.backups#R3, the recovery objectives ofops.backups#R7MUST be written down, and a restore into a separate environment MUST have been tried before hand-over, with the date and the result recorded (ops.backups#R6). - R5. Access. The list of who can reach the system (hosting, the repository, the admin pages, the domain, third-party services) MUST be written down without the secrets. Each account MUST belong to the owner or to a named supporter, and no access MUST depend on a single person who is leaving the project. Secrets shared during the build MUST be changed at hand-over.
- R6. Monitoring. An outside check of the system’s main page and of its most important path MUST run, and a failure MUST reach a named person. Errors MUST be written
to a log that the supporters can read (
ops.backups#R10covers the backup job’s own alert). - R7. Service levels. The service levels MUST be stated, even in small form (
ops.service-levels): what counts as the system working, how fast the supporter reacts, and in which hours. A system without them MUST NOT pass, but the statement MAY be as short as three lines. - R8. Change intake. There MUST be one place where work for the system arrives, and a named person who gives it a class (
process.project-lifecycle#R10). The rules for a live system (process.project-lifecycle#R11,process.project-lifecycle#R12) MUST be known to the supporters, and a staging copy MUST exist or the lack of one MUST be recorded as a risk. - R9. Release and versions. The last release record, the versions of the platform, the runtime and the main dependencies, and the way to deploy and to undo a deploy MUST be recorded.
- R10. Data and security. The personal data the system holds, where it is kept and how long, MUST be listed; the system’s known vulnerabilities and licence obligations MUST be reviewed; and anything that must be renewed (certificates, domains, licences) MUST have its date and its owner.
- R11. Known limits. The risk and technical-debt registers (
arch.risks-and-technical-debt) MUST be handed over with the open items the supporters inherit, each with an owner.
The record and the decision
- R12. The review MUST end in one hand-over record: the date, the reviewers, each line passed or waived, and the decision. A waived line MUST carry the reason, the person who accepted the waiver (the owner for R4, R5, R7 and R10) and the date by which it will be met.
- R13. The lead MUST decide whether the system passes (
process.project-lifecycle#R9). A system with a waiver on R4 MUST NOT pass if it holds data that cannot be recreated, unless the owner records in writing, with the date, that the risk is accepted; the record then says “passed with accepted risk” and names the date by which the restore will be tried. - R14. For a system built for a client, the client MUST receive through the owner the runbook, the access list and the place of the backups, and MUST confirm the hand-over through the owner
(
mgmt.roles-and-decision-rights#R7). The client MUST NOT instruct agents directly. - R15. After a pass, the first change or incident MUST be treated as a check of the hand-over: what the supporter had to ask the builder MUST be added to the runbook within a week.
Guidance
How to run the review
Take one hour per line at most. Sit with someone who did not build the system and ask them to do each thing from the runbook alone: restore last night’s backup into a scratch copy, find the log of yesterday’s error, deploy the previous release. A line that needs the builder to explain is a line that fails. The point of a review that someone else runs is that the author no longer sees what is missing.
Small systems
For a small site or a script, the runbook is one page, monitoring is one external uptime check that mails a named person, and the service levels are three lines. The same lines apply: they get shorter, not absent. A system that nobody supports is a decision to record at R1, not a reason to skip the review.
Hand-over for a live client site
When the work was a rebuild or a migration of a live site, add to R9 the cutover and the rollback that were used, and to R3 the list of the things outside the system that depend on it (addresses, feeds, mail, payment notifications). The delivery rules of the owner’s own support practice ask for the same two things before any work on a live site: a backup proved by restore and a rollback written before the deploy.
Common mistakes
- A runbook that tells the reader what the system is, and not what to do when it fails.
- A backup that exists but was never restored (R4).
- A monitor that mails the builder, who has moved on (R6).
- A waiver with no date: a waiver without a date is a permanent exemption.
- The review done by the builder on the last evening, from memory.
Explanation
Why a review at hand-over. The Google SRE book describes a production readiness review as the step before a team takes responsibility for a service: it checks that the service meets the standards of production setup and operational readiness, and it is a precondition of taking the pager. The same logic holds for a small team: whoever will answer the call must be able to do so without the builder. Why the lines are few. The service manual of the UK government describes the live phase as a time of continued improvement and monitoring; lines that can be tested in an hour are used, long lists are signed and not read. Why waivers carry a date. A hand-over with open lines is normal; a hand-over where nobody knows which lines are open is how a restore is first tried during an outage.
Examples
Good. A small UK online shop built by the owner’s agents is handed to the owner’s support team. The owner records “support team” on 21 November. A reviewer from the support side restores the nightly backup into a scratch copy on 24 November (R4) and finds the uploads folder missing from the backup set; the lead fixes the job and the restore is repeated. The monitor mails the support address (R6), the service levels are three lines (R7), and the hosting login belongs to the owner (R5). The record has one waiver, R10’s certificate renewal date, owner and due 1 December. The lead passes the system. Two weeks later the supporter asks the builder where the mail logs are; the runbook gets the answer within the week (R15).
Bad. The agent that built the site writes “ready for support” in the issue. There is a backup job and nobody has restored it; the monitor mails the agent’s account; the hosting password is in a chat; there is no runbook. The first night the database host is restarted and the site stays down until the owner finds the builder.
Sources
- Site Reliability Engineering, chapter 32 — Google, CC BY-NC-ND 4.0; linked and summarised only, not copied: the production readiness review as the step before an SRE team takes over a service. Accessed 2026-11-21.
- GOV.UK Service Manual, How the live phase works — Open Government Licence v3.0; the live phase as continued improvement and support. Accessed 2026-11-21.
- GOV.UK Service Manual, Monitoring the status of your service — Open Government Licence v3.0; monitoring that reaches a person. Accessed 2026-11-21.
- The delivery rules of the owner’s support practice (a backup proved by restore, a staging copy, a rollback written before the deploy, a closing record), described here as a pattern without client data.