Illustrative review excerpt
Checking an automated merge control
An example of how I'd write up one decision, using findings from my September 10, 2026 Command review. This is my own project, not a client report or a new inspection. The full case study records the broader review and its limits.
Could I rely on this control to check unattended work?
The rules required passing automated checks and a review verdict tied to the exact commit being merged. The inspected inputs were the written rules, workflow list and last run, the merge control's source, and the board's design note, read on September 10.
Two observations changed the answer
- The workflow had been disabled since August 31, while the board deferred merges to it. The documented control wasn't running.
- The control's compiled default accepted an unsigned verdict. Restoring execution alone wouldn't establish who could supply the approval it trusted.
Recommendation
Keep a person responsible for merge approval until the active control has been checked. First choose the path that will enforce approval, then verify what it accepts. Update the rules to describe that path.
Work can continue under that arrangement: an engineer checks the proposed change and its test results, confirms the review applies to that commit, and makes the merge decision. It costs human attention on each merge, but leaves a clear person responsible while the automated path is unverified.
Simply turning the workflow back on is the tempting shortcut. It restores execution without resolving whose approval it accepts. Replacing the whole system would be a larger intervention than these two observations justify. I would first test one chosen approval path and only then decide whether it can take over the repeated check.
Next check and limit
In an isolated test, try a missing verdict, an unsigned verdict, and a valid verdict for an older commit; each should leave the change unmerged. Check the allowed case too: passing automated checks and trusted approval for the current commit.
I would reconsider the human-only recommendation after those cases pass on the selected path, including a check that approval comes from a trusted source, and after the written rules identify that same path. A passing source-level test alone would not establish that the deployed control is running.
These proposed tests have not been run for this sample. The findings don't establish that an unauthorized merge occurred, who could submit a verdict, or whether another control would block it.
Have a technical problem you'd like me to investigate? Tell me what you're working through . We agree on the scope and price first. I investigate, write up my findings, then meet with you to go through the recommendations.