Use this checklist to find where better information or a small change could make delivery easier, safer, or more predictable. Start with a product, service, or project you know and a question you need to answer. Look at recent changes, pipelines, releases, incidents, and the experience of the people involved.
Follow the sections that matter to your question. When an answer is unclear, inspect a real example or ask someone who performs that step. Differences between the records, written process, and people’s accounts can reveal a missing event or a workaround. Make space for individual conversations when people would find it difficult to discuss those differences openly.
Keep the examples and decisions that help you act. An existing issue or team backlog is enough for the next improvement; give it an owner and a point when you'll check whether the result helped.
Where work waits or becomes harder
- Can you follow a small, ordinary change from its request through review, testing, release, and operation?
- Can you explain the longest or least predictable waits along that path?
- Does each person know when work is ready for them and what decision they're expected to make?
- Can you distinguish pipeline queue time, job execution, repeated work, and retries when feedback is slow?
- Can developers get timely help with setup, security findings, access, and release problems?
- Are blocked, abandoned, reopened, and unusually old items visible beside completed-work measures?
- After the last improvement, did the whole delivery path become easier or more predictable?
Trace a recent change and compare its timeline with the path people expected. A long review stage could reflect unclear ownership, a large change, a lack of reviewer capacity, or necessary specialist review. A shorter pipeline won't fix those problems. When one step becomes faster, check the next handoff for a growing queue or extra rework.
Keep quality, security, and stability in the discussion. Some controls take time because they manage a real risk; clearer preparation or a more predictable response may reduce delay without removing the control. A value-stream improvement cycle gives a worked method for tracing a change, choosing a cause to test, and reviewing the result.
Whether the report describes real work
- Can you explain which work and reporting period a measure includes, what starts and ends the calculation, and what it leaves out?
- Does a sample calculated from the underlying records agree closely enough with the report for the decision you're making?
- Are real production releases represented by successful GitLab deployments with the right revision, environment, and completion time?
- Can GitLab associate a merged change with the production deployment used in its lead-time result?
- Do the incident records used for stability measures reflect real production interruptions and service restoration?
- If another tool deploys software or manages incidents, does the information GitLab receives match that tool's records?
- Can you distinguish a delivery change from a change in recording practice, reporting scope, or processing delay?
Check the inputs for the metric you actually use. Deployment frequency needs successful production deployments. Lead time for changes also needs the merge request association. Time to restore service and change failure rate need GitLab incidents. Missing records can make activity disappear; duplicate incidents can increase the reported failure rate.
For Value Stream Analytics, open the items behind a stage result and confirm both events. Work that hasn't reached the end event won't contribute to that stage's duration, so a lower median can coexist with a growing queue. Check units, aggregation, filters, and report processing before comparing periods or teams.
A measure with a known gap may still answer a team's question. Explain the gap beside the result and narrow the conclusion accordingly. If the missing information could reverse the decision, repair the data path first. Reliable DORA and value-stream data covers the deployment, incident, and stage checks in detail.
Whether the measures help teams decide what to change
- Does each measure help someone choose an action that could improve a user, delivery, quality, or risk outcome?
- Is the starting point clear enough to judge whether a change helped?
- Are speed measures reviewed alongside quality, stability, security, cost, and developer effort where those could be affected?
- Can the team explain how someone might improve the number without improving the underlying result?
- Are comparisons limited to work with sufficiently similar definitions, release paths, and recording practices?
- Do the people doing the work help interpret the results and hear what changed in response?
- Have you changed or retired reports that no longer guide a decision?
For example, more deployments may mean smaller, safer releases, but it could also reflect duplicate deployment records. A shorter review stage could mean prompt feedback or review happening outside GitLab. Look at real changes and ask what happened before selecting a response.
Keep activity counts such as commits, lines of code, merge requests, and accepted suggestions out of individual targets. They can describe workload, but don't establish value or explain the constraint. Start with the team's own trend, supported by examples and feedback, before comparing unrelated services.
Choosing measures that guide improvement shows how to select a main result, establish a starting point, and decide what to do when the numbers and people's experience disagree.
Where a feature rollout needs help
- Does the intended task address a problem for the projects and people included in the rollout?
- Is the capability available with the permissions, configuration, integrations, and runner capacity those people need?
- Can someone complete a representative task, recover from a common failure, and find help?
- Are eligible teams using the approach in normal work?
- Has the intended result improved, with acceptable effects on delay, effort, cost, and risk?
- When use is low or the result is disappointing, can you identify which part of the path needs attention?
Consider a security scanner. A successful job shows that the scanner can run. Readiness means someone can assess and route a finding. Use means the workflow handles findings in ordinary work. The team might aim to remediate exposed weaknesses faster without an unmanageable triage queue. Each observation points to different follow-up work.
Check what an adoption report counts before treating it as coverage. DevOps Adoption by group1 can show use of a feature from activity in a single project during the last full calendar month; that doesn't establish use across all eligible projects. Compare the report with representative work and user feedback. Helping teams adopt a new workflow shows how to watch someone complete the task, investigate low use, and choose the next improvement.
Whether improvement work gets followed through
- Does the next improvement describe an observed problem and a desired result clearly enough for someone to act?
- Is there an owner with time and support for the next step?
- Is the proposed trial small enough to learn from while still representing normal work?
- Are dependencies, checks for unwanted effects, and a way to undo or contain the change understood?
- Is there a date or event when you'll compare the result with the starting point?
- Did the last review lead to a decision to keep, change, stop, or expand the approach?
- Can someone find why a valid problem was deferred or what an unsuccessful trial taught?
Choose a manageable improvement with the people who will own it. Consider its likely value, risk, effort, dependencies, and the strength of the examples behind it. An issue can hold the problem, trial, expected result, owner, and follow-up. Include enough detail for someone to continue the work without repeating the investigation.
At follow-up, compare the actual result with the intended outcome. Installing an integration isn't the same as getting complete deployment records, and completing training isn't the same as an easier workflow. If a trial helps, start wider rollout with similar projects and check each group before adding more variation. A value-stream improvement cycle explains how to carry the finding through that decision.
Further Reading
Notes
-
DevOps Adoption by group is available in Ultimate. You can also review activity directly in the projects included in a rollout. ↩
- Value Stream Analytics explains how to select stage events and inspect the work behind a duration.
- DORA metrics explains the records, calculations, and assumptions to check when a delivery or stability result is unclear.
- DevOps Adoption by group explains what the adoption indicators include and when they're updated.
- Description templates helps you create a reusable starting point for improvement issues.