Lesson 5 of 5
Gate policy
17 min
Listen to this lesson
A gate is only as good as the rules behind it. Too loose, and it stops nothing. Too strict, and teams find ways around it, which is worse than having no gate, because people believe they are protected. A gate policy states, for each gate, what must be true for a change to pass, who can grant an exception, and how exceptions are recorded and reviewed.
Elements of a gate policy
| Element | Question it answers | Contoso example |
|---|---|---|
| Scope | Which changes does the gate apply to? | All workbench transports to CMQ; all builds of the warranty portal |
| Criteria | What must be true to pass? | No ATC priority 1–2 findings; all ABAP Unit tests pass; pipeline unit and acceptance stages green |
| Thresholds | How much is tolerated? | Regression in CMQ: 100% of critical process tests pass; at most 2% non-critical failures with analysis |
| Exceptions | Who can override, and how? | Kenji for ATC exemptions; for regression failures, the process owner approves the exception and Priya (test lead) records the risk and the expiry |
| Records | How is the decision kept? | Exemption with reason and expiry; signed decision on the feature |
| Review | When is the policy checked? | Monthly gate metrics; policy review each release |
The parts of a gate policy.
Setting thresholds
Thresholds should follow risk, not convenience. Critical processes, such as order to cash for Great Lakes Fleet Services, procure to pay with purchase order approval over $25,000, and period-end close, get zero tolerance. Lower-risk areas can tolerate a small number of known, analyzed failures. Percentages alone are dangerous: "95% passed" can hide the one failure that matters. Express thresholds by risk class, and require that every failure, even a tolerated one, has a classification and an owner.
Exceptions that keep the gate credible
Every real gate needs an exception path, because sometimes the business must ship with a known issue. What matters is that the exception is visible and accountable. A good exception has an accountable approver who owns the risk, a written reason, a scope (which finding or test, which change), an expiry date or a linked fix, and a record that auditors and later teams can find. ATC's exemption process already provides this for static findings; for regression failures, the decision belongs on the feature or release record in Cloud ALM.
- Is the failing check related to the change, or to a known issue?Related → fix before release · Known issue → continue
- Does the failure affect a critical process or control?Yes → no exception without process owner approval; Priya records the risk and the expiry · No → continue
- Is there a workaround and a planned fix with a date?Yes → exception possible, recorded with reason and expiry · No → do not release
- Is the exception recorded where auditors and the next team will find it?Yes → proceed · No → record it first
Measuring the gates
- Block rate: share of changes stopped at each gate. Very low may mean the gate checks too little; very high may mean checks are noisy or badly targeted.
- Escaped defects: defects found later that the gate could have caught. The best measure of gate value.
- Exception count and age: how many exceptions are open and how many have passed their expiry.
- Time in gate: how long changes wait; slow gates encourage bypassing.
- False block rate: blocks caused by flaky tests or wrong checks; these erode trust fastest.
What would you do?
Two days before a release, a regression test for incoming payment processing fails because of a defect in a rarely used cash discount variant. The fix will take a week. Business users have a manual workaround. What should the gate policy allow?
- A gate policy sets scope, criteria, thresholds, exceptions, records and review for each gate.
- Set thresholds by risk class; critical processes get zero tolerance; every failure needs a verdict.
- Exceptions need an accountable approver, reason, scope, expiry and a findable record.
- Emergency changes use a fast gate, not no gate, followed by full regression.
- Measure block rate, escaped defects, exception age, time in gate and false blocks.
Check your understanding · not graded
Lesson check
1. What must a gate exception include to keep the gate credible?
2. Why is "95% of tests passed" a weak gate threshold on its own?
3. Which metric best shows whether a gate is adding value?
4. An emergency fix must reach CMP tonight. What should the gate policy require?