Table of contents
Research context
This protocol measures the share of offshore releases that keep a rebuildable build artifact. The unit is one released version on a named environment inside the observation window. A release counts as rebuildable when a retained artifact can be deployed again and the recorded commit can reproduce it. The study describes release evidence, not the quality of the code in the release.
Define the release window and the environments in scope. Record the repository, the pipeline, and the deployment inventory used to build the list. If the pipeline changes during the window, split the sample at the change date. A new pipeline can change the artifact format and the result.
Define an artifact by its digest and its type, such as a container image, a static bundle, or a signed mobile package. Record the commit, the pipeline run, the build time, and the deploy time for each release. Store the digest rather than a mutable tag. Test the definition on the first five releases.
Observe the artifact store rather than the release notes alone. Look for a release that is deployed but has no retained artifact, and a retained artifact that no release references. Record the discovery source for each mismatch. The study should report the orphans as well as the gaps.
Test a rebuild from one retained artifact per release group. Deploy it into an isolated environment and confirm that it starts and serves the expected version. Record the time taken and the manual steps needed. A rebuild that needs tribal knowledge is a partial result, not a pass.
Practical application
Check whether the recorded commit still builds from a clean checkout with the locked dependencies. Compare the new digest with the retained digest where the build is reproducible. Record a difference and investigate it before calling it harmless. Report the releases that could not be tested and why.
Report the number of releases, the number with a retained artifact, the number that redeployed, and the number that rebuilt from source. Show the counts by environment and by artifact type. Present the releases with no record and the releases with an expired artifact separately. Use the same denominator for every rate.
Separate a missing artifact from an expired one. A missing artifact was never retained, while an expired artifact was retained and then removed under a retention rule. Report the two categories separately because the fixes differ. Record the retention date and the deletion ticket beside each expired case.
Inspect the releases around a pipeline migration or a storage change. A cluster of gaps may mean the new pipeline publishes to a different bucket that the index does not read. Compare the pattern with the change log. A single gap can be a mistake, while a cluster is a design gap.
Do not treat a retained artifact as proof that the release can be rebuilt. The artifact may start while the recorded commit no longer builds because a dependency disappeared. Sample one release and attempt the source rebuild. Report what the attempt showed.
Operating controls
The record cannot show whether the deployed build is the build that was approved or whether it was tampered with after publication. It can show only whether an artifact exists and whether one rebuild succeeded. A missing artifact is not proof of a bad release, and a present artifact is not proof of provenance. The protocol also cannot see an artifact that was deployed by a vendor outside the pipeline.
A useful additional view is the number of releases whose artifact digest matches the digest recorded at deploy time. That match is a simple provenance check. Record the deploy-time digest and the stored digest for each release. That field turns a count into a review of integrity.
Give the result to the client technical owner and the offshore delivery lead with one observable change. The change might be a pipeline publish step, a retention period, a storage owner, or a quarterly redeploy test. Record the owner, the effective date, and the measure that will show whether the change worked. Repeat the protocol with the same definitions after the change.
Close the review with one change and a date to re-measure. The change should address the largest missing or expired group, not the easiest release to test. Record the expected effect and the evidence that will prove it. A retention rule that is never tested is a storage setting, not a control.
Sources
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.
