Developer Offshore logoDeveloper OffshorePowered by StealthAgentsScope a developer
Table of contents

Developer Offshore research

Time zone overlap hours and review latency in offshore engineering

A research review of overlap hours in offshore engineering: how to define usable overlap, where review latency accumulates, what to measure, and a four-week review protocol.

Use this report with the Research library and related developer guides to turn evidence into a bounded work brief.

Time zone overlap hours and review latency in offshore engineering

Key stats

  • 3 latency segments: queue wait, context time, and decision time
  • 2 overlap measures: nominal hours and quiet hours available for review
  • 2 timestamps per request: opened locally and first substantive response
  • 1 four-week experiment at a time

Key takeaways

  • Measure quiet overlap rather than clock overlap, because meetings consume the shared hours.
  • Separate queue wait from review work so the clock is not blamed for a routing problem.
  • Track the share of reviews completed in one overlap window and inspect the slow group.
  • Keep a named primary and backup reviewer so a request does not wait for one calendar.
  • Change one scheduling or routing variable per four-week experiment.

Question and scope

This report examines how time zone overlap affects review latency in offshore software delivery. The unit is one review request and the elapsed time until a reviewer provides a decision or a substantive comment. The scope covers code review, design review, and release approval handled by a buyer-side reviewer and an offshore pod. It excludes customer support tickets, which have different service targets and routing.

The question is whether teams treat overlap hours as a scheduling fact or as a measurable constraint on flow. A calendar may show four shared hours while meetings consume most of them. The report looks for the difference between nominal overlap and time actually available for review. It treats that difference as a planning input rather than a personal complaint.

Define overlap precisely

Nominal overlap is the clock time when both sides are online. Usable overlap is the portion that remains after recurring meetings, standups, and handoffs. A team with four shared hours may have less than one hour of quiet review time. Record the calendar blocks that remain open before calling the overlap sufficient. The distinction matters because adding a meeting silently reduces review capacity.

Overlap also varies by role. A reviewer who also manages incidents may lose the shared hour without warning, while an engineer with a protected block keeps it. Map the overlap for the reviewer role rather than for the team average. If only one person can approve a change, the effective overlap is that person's calendar, not the group's. Map the reviewer calendar before agreeing to a service target that assumes a shared hour exists.

Where latency accumulates

The first delay is queue wait, when a request sits unassigned because the reviewer is offline. The second is context time, when the reviewer reads the change and reconstructs the intent. The third is decision time, when the reviewer waits for a test result or a product answer. Only the second and third are visible inside the review tool, so queue wait often goes unrecorded.

A request opened at the end of the buyer's day can wait a full offshore shift before anyone reads it. The author may then be offline when the comment arrives, adding another cycle. That pattern produces a two-day review that contains only twenty minutes of actual work. Measuring the request timestamp and the first substantive response exposes the gap. Report the request time in both local zones so the reader does not have to convert it by hand.

Review routing and queue age

Route reviews to a small group with a named primary and backup, not to the whole team. A broad reviewer pool looks flexible but often produces diffusion, where each person assumes someone else will pick it up. Set a queue-age threshold that triggers a reassignment. The threshold should be visible to both the author and the reviewer, so the handoff is not a private message.

Prefer smaller changes that can be reviewed inside one overlap window. A large change that needs a scheduled walkthrough will usually wait for the next day. Split it by behavior or by layer so the reviewer can approve one part while the rest is still in progress. The split also makes the review evidence easier to attach to a release.

Meeting design versus written review

A meeting can resolve a disputed design faster than a comment thread, but it consumes the scarce shared hour. Reserve live time for decisions that need negotiation, and move status updates to writing. A written summary with the decision and the owner lets the offshore team act during its own hours. The meeting then becomes a confirmation rather than the only place where context exists.

Record the outcome of each live review in the pull request or the design document. A decision that lives only in a call is invisible to the engineer who starts work after the call ends. Include the alternatives that were rejected and the reason, so the next reviewer does not reopen them. The record should be short enough to read before the next overlap window.

What to measure

Measure queue wait, first substantive response, and total review time as separate values. Add the local hour when the request was opened and when it was answered. That pair shows whether the delay comes from the clock or from the review process. Report the median and the oldest open request rather than an average that hides the tail. A combined figure hides whether the team should change routing or reduce the change size.

Track the share of requests that complete within one overlap window. A low share does not prove a time zone problem, but it identifies where to look. Compare the result with the size of the change and the number of reviewers required. A small change that still misses the window is a routing issue more often than a scheduling one.

A four-week review protocol

Run the measurement for four weeks without changing the schedule. Tag each review with its size, its required approvers, and whether a live discussion happened. At the end, group the requests by whether they completed in one window. Look for the common condition in the slow group rather than debating anecdotes. Leave the review rules untouched during the baseline so the result reflects the current process.

Choose one change for the next four weeks, such as a protected review hour or a smaller change limit. Re-measure the same values and compare the result with the baseline. If the change helps, keep it and document the new expectation. If it does not, try one different change rather than adding several at once. Write the expected effect before the experiment so the result can be judged against a stated hypothesis.

Limits of the signal

Review latency is only one part of delivery time. A fast review can still sit behind a slow test suite or a release freeze. The measure should be read beside build time and deployment frequency. On its own, it can reward reviewers who approve quickly without reading, which is worse than a slow review. Report the review measure beside build time and release wait so the reader sees the whole path.

The signal also depends on how the team records timestamps. Comments that only say looking can reset a queue-age clock without providing a decision. Define what counts as a substantive response before using the measure. The definition should be written down and applied consistently across both sides. Publish the definition in the working agreement and apply it to both sides of the review.

From latency to a working agreement

Use the review result to write a small working agreement between the buyer and the pod. Name the quiet overlap hours, the review size limit, and the response expectation for a normal change. Add the escalation path for a release-blocking review that misses the window. The agreement should fit on one page so both sides can apply it without interpretation.

Review the agreement after each quarter and change one clause at a time. A clause that is never used may be unnecessary, while a clause that is constantly waived points to a wrong expectation. Record the waivers so the pattern is visible to both sides. The agreement is a coordination tool, not a service-level threat. Keep the revision history beside the agreement so both sides can see what changed.

Reporting the result

Report the overlap hours, the review latency trend, and the share completed in one window. Add the one change the team is testing and the date of the next comparison. Keep the detail in an appendix for the people who want to inspect individual requests. The buyer should see the constraint and the experiment, not a claim that the time zone is the only cause.

Review the result at the next service meeting and decide whether to keep the change. If the latency persists, examine the approval rules and the test feedback loop before adding more meetings. A calendar change is cheap to try and easy to reverse, which makes it a good first experiment. The record of that experiment is more useful than a general agreement to communicate better. Record the change in the working agreement so the next pod inherits the lesson instead of repeating it.

Evidence table

Signals, inspection points, and accountable owners
SignalWhat to inspectOwner
OutcomeAcceptance evidence for the bounded taskTask reviewer
ControlAccess, test, and approval boundaryInternal owner
HandoffOpen risks and next decisionNext owner
Good distributed work is observable at the handoff: the result, evidence, limitations, and next owner are explicit.

Frequently asked questions

Is more overlap always better?

Not necessarily. More shared hours can turn into more meetings, which reduces quiet review time. Measure the usable hours and the review result before extending the schedule.

Should we count comments that only acknowledge a request?

No. A substantive response contains a decision, a question that blocks progress, or a requested change. Acknowledgement alone should not stop the queue-age clock.

What if the buyer has only one qualified reviewer?

That person's calendar is the real overlap. Name a backup with the same authority or reduce the approval requirement for low-risk changes.

Sources

  1. GitHub Docs, About pull request reviews
  2. Google SRE Workbook, Postmortem Culture
  3. NIST SP 800-128, Guide for Security-Focused Configuration Management

Related research