Find Where Your Delivery Pipeline Gets Stuck
Find Where Your Delivery Pipeline Gets Stuck
Run a weekly review of where pull requests wait. In Weave Engineering Intelligence, start with review volume, rounds before merge, and PR deploy lead time. Then open the records behind the slowest interval. Each interval points to a different change your team can try.
Compare three intervals in the same set of repositories
| Interval | What to inspect | One change to try |
|---|---|---|
| PR opened to first review | Open PRs awaiting a response, reviewer workload, and when the change became ready for review | Rotate first-review coverage when the queue builds up around one reviewer |
| First review to merge | Review rounds, the changes requested, and the scope of the PR | Split changes when large PRs repeatedly require several rounds |
| Merge to production deploy | PR deploy lead time, deploy frequency, and failed deploys | Check whether batching, a release window, or failed builds explain the gap |
Use the same repository list and time window each week. Track open PRs alongside merged ones so changes still stuck in review appear in the discussion. If deploy records are missing for some repositories, report which repositories have usable coverage before comparing them.
Open the PRs behind the slowest stage
Compare a few slow PRs with a few typical ones. A large migration may take several rounds because the work is complex. A routine dependency update sitting untouched for two days points to review availability. Weave's code output measurement estimates expert effort from the changes in a merged PR and shows the reasoning behind the estimate. Use that alongside change size and review context when deciding which items to compare.
Check for a trade-off the following week
Record the change you tried and compare the same stage in the next comparable period. If first reviews happen sooner, check review cycles and reverts alongside wait time. Faster pickup has limited value if the team needs more rounds to get the code ready. Write down any change in staffing, repository scope, or release cadence that could explain the difference.
For a first pass, take one repository with a recurring backlog and review the last two weeks of PRs. Identify the longest wait, open three representative records, and assign one improvement to an owner.