Key takeaways
- Write done as observable evidence, not as an activity the team performed.
- Separate merged from done so a review comment is not mistaken for completion.
- State the test, review, and documentation evidence each task must carry.
- Allow exceptions through one written route with a named approver.
- Audit a small sample each month and remove criteria that never change a decision.
Start with observable evidence
A definition of done should describe what a reviewer can see, not what the author intends. Replace the item tests written with a link to the passing test run and the case it covers. Replace documentation updated with the changed section and the date it changed. Observable evidence lets someone in another time zone verify the claim without a conversation. It also makes a disagreement about done a discussion about facts rather than effort.
Write the list with the people who review the work, including the offshore tech lead and the buyer owner. A definition invented by one side tends to include criteria the other cannot check. Keep the list to the items that actually change a release decision. A long checklist invites people to tick boxes without reading them. Six to ten specific criteria are usually enough for a delivery pod.
Separate done from merged
Merged means the change is in the shared branch, which is not the same as ready for customers. A feature can be merged behind a flag, missing a migration, or waiting on a content update. The definition should say which state counts as done for the task type. That distinction stops a status report from claiming completion when only the code has landed. It also tells support whether they can describe the change to a customer.
Use separate status values for in review, merged, verified, and released where the work needs them. Not every task needs all four, but the team should agree which apply. A bug fix might be done when the fix is verified in staging, while a schema change is done only after the rollback path is tested. Matching the definition to the task type is more useful than one universal checklist. The point is a shared meaning, not more columns on a board.
Cover tests and review
State the test evidence each change must carry. For a small fix that might be one regression test and the output of the relevant suite. For a risky change it might include a negative case, a manual check, and a note about the data used. The definition should name the level of evidence rather than demand full coverage for every ticket. Reviewers can then reject a change for a missing case without arguing about general quality.
Review evidence belongs in the definition too. A named reviewer, the decision, and the follow-up items are part of done for most work. The author should link the review rather than describe it from memory. If a review produced a follow-up, the follow-up needs an owner and a date or an explicit decision to drop it. Otherwise done becomes a moment when problems are quietly transferred to nobody.
Handle exceptions explicitly
Some work cannot meet every criterion, and the definition should say so. A hotfix might ship before the documentation update, and a migration might need a manual step during a freeze. Allow an exception through one route with a named approver and a short reason. Record the exception in the ticket rather than in a chat message. The record lets the next reviewer see that the gap was accepted rather than missed.
Review exceptions each month and look for a pattern. If the same criterion is waived repeatedly, it is either unrealistic or the process around it is broken. Change the criterion or fix the cause, but do not leave a rule that everyone quietly ignores. A definition of done loses value the moment it becomes a document people work around. The exception log is the signal that keeps it honest.
Publish it where work starts
Put the definition where a ticket is created, not in a folder three clicks away. A short block in the ticket template or the board header is more effective than a long policy page. The developer should see the evidence requirements while writing the estimate. Reviewers should be able to link the definition when they return a change. If the text is hard to find, it will not shape the work.
Use the same words the team uses in review. If everyone says proof instead of evidence, the definition should use proof and explain what counts. Add one good example for a small task and one for a risky task. Examples resolve more ambiguity than another paragraph of rules. Keep the definition short enough to read in under a minute. A definition that is easy to find is used more often than one that is merely correct.
Audit a sample each month
Sample a handful of completed tickets and check whether the claimed evidence exists. Look for the test run, the review link, and the documentation change rather than trusting the status. If a sample fails, the question is whether the criterion was unclear, the tooling made it hard, or the team skipped it under pressure. Each answer leads to a different fix. The audit should take less than an hour and produce one concrete change.
Share the result with the buyer and the provider lead without naming individuals. The goal is a definition that the pod can actually meet on a normal week. If the sample always passes but releases still surprise the buyer, the criteria are measuring the wrong thing. Add the missing evidence and remove the items that never affect a decision. A definition of done is useful only while it predicts what the buyer will receive.
Questions about assessing Philippine developers
How many items should a definition of done have?
Six to ten that change a release decision. More than that usually produces box ticking rather than better evidence.
Should every task use the same definition?
No. Keep shared criteria and add the evidence a task type needs, such as a migration check or a manual accessibility pass.
Who owns the definition of done?
The buyer owner and the offshore tech lead share it. Reviewers apply it, and the monthly sample shows whether it still fits.
