{"canonical_url":"https://engineer.company/glossary/definition-of-done/","content_markdown":"# Definition of done\n\n## What it is\n\nA Definition of Done sets the common quality conditions work must meet before the team regards it\nas part of a completed product increment. “It runs on my laptop” identifies one observation. A\nDefinition of Done gives everyone a way to tell when the same work is integrated, verified at the\nappropriate level and ready for another person to use or maintain. Its purpose is to make the\nstate of the product visible.\n\nThe Scrum Guide calls the Definition of Done a formal description of the Increment when it meets\nthe product's required quality measures. Work that falls short cannot be released or presented as\npart of the Increment at the Sprint Review. It returns to the Product Backlog for future\nconsideration. If the organization sets a standard, Scrum teams follow it as a minimum. Teams\nworking together on one product use a common Definition of Done. Two development groups can still\nuse different tools to collect the evidence.\n\nAcceptance criteria name what a particular change must do. A change to an export, for instance,\ncan require that archived accounts appear only when requested. A Definition of Done can require\nthat changes with persistent data are integrated, verified on representative existing records and\naccompanied by updated operating instructions when those instructions are affected. The export\ncriteria describe its particular behavior. The completion standard governs the quality of work\nacross the product. Both claims need evidence when the export changes.\n\n## Why it matters\n\nA shared finish line reduces uncertainty in planning and handover. Suppose a background job has\npassed unit tests but no one has run the migration against older records. The author may view the\nitem as effectively complete, while the operator sees an unresolved release risk. A completion\nstandard states what needs to happen before another team plans around the change. It also gives\npeople a straightforward way to say which evidence is missing.\n\nSelect conditions based on the failure modes the team repeatedly needs to manage. A library might\nrequire tests of changed public behavior, compatibility evidence for affected versions and a\nclear upgrade note for a breaking change. A service that processes personal information may need\na review of how each changed data path handles that information. An interface team might have\nshared keyboard requirements and a check of the changed text in its supported editions. The\nconditions describe a state that can be inspected. The command, artifact or review that\ndemonstrates each state can be recorded separately.\n\nThe Definition of Done also makes comparisons meaningful. If one team calls an item complete at\ncode review and another waits until validation on representative data, counting both items under\nthe same status obscures the amount of work left. A common standard does not make two features\nequally hard. It lets the people using the status understand what still needs to happen before\nrelying on the result.\n\n## What it looks like in practice\n\nConsider a fictional reporting product. It loads a supplier's file overnight and keeps the\npreviously verified report available when an import fails. The next change accepts a new optional\nfield. Its acceptance criteria cover the new field and what readers see when the field is\nmissing. The shared Definition of Done for import changes might include the following conditions.\n\n- The change is integrated into the product version that will be evaluated as a whole. The\n  acceptance examples and affected existing behavior are verified with representative input.\n- A retry and a partially completed import are exercised in a safe test environment. The previous\n  verified report stays available when the new import cannot finish, and a retry does not\n  duplicate data.\n- Any changed operator procedure is updated where the operator will look for it. The release\n  record identifies the test inputs, outcomes and any limitations that matter to the team.\n\nUse a real candidate build in this example, not a sequence of screenshots of unconnected parts.\nRun the job once with an existing file, once with the new field and once with a controlled\nfailure. Record the report identifier before and after the failure. Run the retry and compare\ntotals and row identities. Ask another operator to find the revised instructions and identify the\navailable report. That quick check can catch a useful procedure placed where no one will find it.\nIf a test cannot observe the affected state, mark the evidence incomplete and fix the\nverification.\n\nOur [service levels page](/service-levels/) describes documentation and a handover covering code,\naccess, credentials and current status. These are examples of results someone can inspect. A team\ncould explicitly require a handover note that another maintainer can use, then ask that person to\nshow where they would find a relevant procedure or open issue. This illustrates one possible\ncompletion condition. The page describes our published commitments, independent of any team's\nchoice of process.\n\n## Where it goes wrong\n\nProblems start when the wording cannot settle a decision. “Test thoroughly” offers little help in\ndiscussing the stalled import. “Verify recovery once” leaves room to stop before the first report\nbecomes readable again. State which relevant behavior has to be observed and what would show that\nit still works. For work to which a check genuinely does not apply, call out the reason and\npreserve the checks that do apply. Changing the completion status should not require inventing an\nexception from scratch.\n\nAnother trap is turning the Definition of Done into a long task inventory. A shared condition\nstates what must be true. The route to that state depends on the change. One team checks\nrecoverability with a fixture containing valid old reports, while another must rehearse a restore\nfrom an actual backup. The evidence differs. The condition about recoverability remains\nunderstandable to both groups. Keep instructions near the tools, and update the common conditions\nwhen experience shows that an important state is still unclear.\n\nFinally, a successful test suite can establish only its assertions. A test might verify that a\nreport exists and still fail to notice that the report is yesterday's. Ask what each signal\nproves and whether the evidence came from the same candidate that would advance. If an item\ncannot meet the standard this time, record the missing evidence and the next required work.\nPeople can then distinguish useful progress from an increment ready to rely on.\n\n## Further reading\n\n- [Scrum Guide on the Definition of\n  Done](https://scrumguides.org/scrum-guide.html#commitment-definition-of-done)\n- [Scrum.org's guidance on the Definition of\n  Done](https://www.scrum.org/resources/what-definition-done)\n- [UK Government Service Manual on user stories and acceptance\n  criteria](https://www.gov.uk/service-manual/agile-delivery/writing-user-stories)\n\n\u003chttps://engineer.company/glossary/definition-of-done/\u003e\n","date_modified":"2026-11-19T19:18:49Z","description":"How a Definition of Done makes completion observable across a product, with a batch import example and a way to handle work that falls short.","formats":{"gemtext":"https://engineer.company/glossary/definition-of-done/index.gmi","gopher":"https://engineer.company/glossary/definition-of-done/gophermap","html":"https://engineer.company/glossary/definition-of-done/","json":"https://engineer.company/glossary/definition-of-done/index.json","markdown":"https://engineer.company/glossary/definition-of-done/index.md","pdf":"https://engineer.company/glossary/definition-of-done/index.pdf"},"kind":"page","language":"en","title":"What is a definition of done?","url":"https://engineer.company/glossary/definition-of-done/","version":1}