Outsourced ProgrammersPowered by StealthAgents

What share of offshore pull requests are abandoned before merge?

A measurement protocol for pull requests that close without merging, with reasons, age, and the review stages they reached.

What share of offshore pull requests are abandoned before merge?
WORK_LANE / RESEARCH BRIEF / DELIVERYREVIEW QUEUEACCEPTANCEPASSowner review requiredEVIDENCE TRAILTICKET → BRANCH → REVIEW → CLIENT RELEASE
Table of contents
  1. Research context
  2. Practical application
  3. Operating controls
  4. Sources
  5. Frequently asked questions

Research context

This protocol asks what share of offshore pull requests are abandoned before merge and why. The unit is one pull request opened by an offshore engineer during the window, whether it was merged, closed without merge, or left open past the observation date. The study measures the share closed without merge, the review stage reached, and the reason recorded. It does not evaluate the engineer's performance or the wisdom of every rejection.

A closed pull request is not automatically a failure. It may have prevented a risky merge, duplicated a change that shipped elsewhere, or been replaced by a smaller branch. The protocol should preserve that possibility instead of treating every closure as waste. The close reason is therefore as important as the count.

Set the scope to one repository, one review policy, and one offshore team. Record the branch protection rules, the required reviewers, and the definition of an abandoned pull request. If the policy changes during the window, keep the earlier definition for the full sample or analyze the change as a separate period. A new required check can look like more abandonment when it is only a different gate.

Measure the age at closure in business hours rather than calendar days when the team works across time zones. A branch closed after a two-week holiday may have been active for only a few working days. Report both if the difference is large. The reader should see the queue experience, not just the timestamp difference.

Define abandonment as a pull request closed without merge, or open for more than the agreed review window with no activity and no owner. A pull request that is replaced by a smaller successor is one unit linked to the successor, not two independent failures. A draft that never reached review is excluded and counted separately. Each definition needs a positive and a boundary example so two reviewers classify the same case the same way.

Record whether the author was still assigned to the client when the pull request closed. A change abandoned after a reassignment is a staffing event, not necessarily a review failure. If the author left, name the replacement owner and whether the work was transferred. That context prevents a misleading abandonment rate.

Record the author, the change type, the size band, the risk tier, the reviewers, the first review time, the last activity time, and the close reason. Add whether the change was superseded, deferred, duplicated, or rejected on technical grounds. Keep the pull request history as it existed at the observation date. A second reviewer should be able to repeat the classification from the record.

Separate bot-generated pull requests from human ones. Dependency bots can open and close many branches that say nothing about the team's review behaviour. If bots are included, report them as a separate category and exclude them from the headline rate. The same rule applies to generated code and lockfile-only changes.

Practical application

Report the number of pull requests, the share closed without merge, the share still open past the window, and the share superseded. Show the counts by change type, size band, and author. Present the median age at closure and the review stage reached. Use the same denominator for every rate and show the number of records with missing close reasons.

Look at the review comments on abandoned pull requests to see whether the author received a clear next step. A closure after a vague comment may indicate a process gap rather than a decision to stop. Record whether the last comment named an action, a question, or a rejection. That field turns a count into a diagnosis.

Inspect the abandoned pull requests and group the reasons without naming the engineer. A cluster of stale branches may point to a queue that nobody owns, while a cluster of rejected changes may point to an unclear requirement. Compare the result with the review latency and the number of reviewers. Report drafts and bot-generated pull requests separately because they carry different intent.

Track the share of abandoned pull requests that were later reopened or recreated. A reopened change suggests the work was still needed and the closure was procedural. A change never recreated may have been genuinely obsolete. The two cases should be reported separately because they imply different fixes.

The record cannot show whether the author lost interest, was reassigned, or received a verbal instruction to stop. It can show only that the pull request closed or stalled. Abandonment is not proof of waste, because a closed change can prevent a worse merge. The protocol also cannot capture work that was moved to another branch without a link.

Do not use abandonment as a individual performance metric. A team that discourages closing bad work will accumulate stale branches and call it productivity. The measure is useful at the queue and policy level, where it can reveal unclear ownership or an unrealistic review limit. Keep the individual data out of the summary.

Give the result to the engineering lead and the client owner and ask them to choose one observable change. The change might be a stale-branch sweep, a required close reason, a smaller review limit, or a weekly queue review. Record the owner, the effective date, and the measure that will show whether the change worked. Repeat the protocol with the same definitions and compare the same fields.

Ask two reviewers to classify a sample of closed pull requests independently and compare their labels. A low agreement rate means the definitions are still ambiguous. Revise the examples before collecting the full sample. The exercise is cheap and prevents a confident report built on inconsistent categories.

Operating controls

A useful additional view is the time from the last review comment to closure. A branch that sits for weeks after a clear rejection is a queue hygiene problem, not a review problem. Report the median and the longest cases. The owner can then decide whether a stale-branch sweep is worth the effort.

End with one policy change and a date to re-measure. The change might be a required close reason, a weekly stale-branch review, or a smaller default pull request size. Record the owner and the expected effect. Repeat the protocol with the same definitions so the comparison is honest.

Record the pull request size and the number of files changed. A large branch is more likely to be abandoned because the review cannot finish before the work becomes stale. If the largest branches dominate the abandoned set, the fix may be a smaller default change size rather than more reviewer discipline. Report the size distribution beside the closure rate.

Check whether the branch was based on an old main and never rebased. A branch that conflicts with months of changes may be closed because merging it would be riskier than rewriting it. Record the age of the base commit at closure. That field separates a process problem from a technical one.

Ask the author, when available, whether the closed work was replaced or dropped. A short note turns a silent closure into a usable lesson. If the author has left the engagement, ask the reviewer who handled the closure. The goal is to understand the decision, not to assign blame for a branch that no longer mattered.

Report the number of open pull requests at the end of the window as a separate queue measure. A team can close few branches and still carry a large review backlog, which the abandonment rate alone will hide. Show the open count, the age of the oldest branch, and the number with no reviewer assigned. That view tells the owner whether closure is the right problem to solve first.

Ask whether the repository has a rule for stale branches and whether anyone enforces it. A written rule with no owner is only a suggestion, and a bot that closes branches without a reason can hide useful work. Record the rule, the owner, and the last enforcement date. If no rule exists, propose one with a grace period and a clear notification before closure.

Sources

  1. GitHub Docs, about pull requests
  2. Google Engineering Practices, code review
  3. Open Source Guides, contributing guidance
  4. NIST Secure Software Development Framework

Frequently asked questions

What should happen first?

Begin with a bounded ticket, approved access, explicit acceptance evidence, and a named reviewer.

Who approves production changes?

The client's technical owner keeps final merge and release authority.

Related research

How long does a new offshore engineer take to make a first production change?

What share of offshore defects are found after release?

Build a programming lane your technical owner can review.

Bring the stack, backlog, acceptance checks, and release boundaries. We will help scope the role around the work.

Discuss your programming needs