The path a change takes
- Ticket: someone records why the change is needed, in Jira or a similar tool.
- Branch: an engineer makes the change in a separate copy of the code.
- Pull request (PR): they ask for it to be merged into the main code.
- Review and approval: another engineer reviews and approves it.
- CI: automated builds and tests run, including security checks.
- Merge and CD: once approved and passing, the pipeline deploys it to production.
Environments
Changes are tried in development and test (often called staging) before production, where real customers and data are. Keeping these separate is a control (ISO A.8.31). Developers with standing write access to production can skip the whole process.
Where it goes wrong
- Self-approval: the author approves their own change.
- Approval after deployment: the approval is real, but too late to count.
- Admins bypassing branch protection, the setting that forces reviews.
- Emergency changes with no review afterwards.
What this means for evidence
- The branch protection setting requiring review, with no bypass for admins.
- For each sampled change, the PR showing a different approver, the approval time before the merge time, and the linked ticket.
- The deployment log, which gives the full list of changes, including any made without a PR.
Branch protection rule: main
- Require a pull request before mergingOn2
- Required approvals13
- Dismiss stale approvals when new commits are pushedOn
- Require status checks to passOn: build, tests, snyk
- Do not allow bypassing the above settingsOn (includes administrators)4
- 1Date and repository.
- 2Pull request required, so every change is visible.
- 3At least one approval by someone other than the author.
- 4No bypass, even for admins. Without this the rule is optional.
Weak version: Administrators can bypass the rule, and nothing stops them merging directly.
#1482 Rate limit the export endpoint
- Author
- imran.s1
- Approved by
- david.e on 2026-09-17 14:222
- Merged
- 2026-09-17 15:01 by imran.s3
- Linked ticket
- CHG-221: Export endpoint abuse4
- Checks
- build passed, tests passed, snyk passed
- 1The author.
- 2A different person approved it, and the time is recorded.
- 3Merged after the approval. Compare the two timestamps on every sample.
- 4Linked to a ticket that says why the change was needed.
Weak version: Approved after it was merged. An approval after the fact proves nothing about the change before it went live.
Production deployments, September 2026
| Deployed | Commit | PR2 | Triggered by | Environment approval3 |
|---|---|---|---|---|
| 2026-09-17 15:09 | a41f9c2 | #1482 | merge to main | david.e |
| 2026-09-22 11:30 | c07e118 | #1490 | merge to main | maria.s |
| 2026-09-28 02:14 | e9d3a70 | none (hotfix) | manual | david.e4 |
- 1The whole period. The auditor needs the complete population before sampling.
- 2Each deployment ties back to a PR.
- 3Environment approval before production.
- 4An emergency change. The auditor will ask for its after-the-fact review under the emergency change procedure.
Weak version: A list of merged PRs instead of deployments. It misses anything deployed without a PR, which is exactly what the auditor is looking for.
Ticket, pull request, someone else approves, pipeline deploys. That chain is the change control.
1. An engineer approves their own pull request. What is wrong?
2. Which list should an auditor sample changes from?
3. Approval at 15:10, merge at 14:32. Does it hold up?