Make feature flag targeting rules disagree on purpose , September 11 review

A September 11 test of overlapping feature-flag rules, documenting which identity, group, percentage, and default target wins for each persona.

Make feature flag targeting rules disagree on purpose , September 11 review
WORK_LANE / FIELD GUIDE / QUALITYREVIEW QUEUEACCEPTANCEPASSowner review requiredEVIDENCE TRAILTICKET → BRANCH → REVIEW → CLIENT RELEASE
Table of contents
  1. Quick read
  2. Readiness check
  3. Reproduce the feature flag precedence failure safely
  4. Leave enough evidence for a second person , September 11 review
  5. Role brief
  6. Frequently asked questions
  7. Sources

Is the handoff ready?

Use these checks before access expands or the first ticket begins.

Question

Ready: One observable feature flag precedence outcome

Needs work: A request to inspect the whole system

Fixture

Ready: synthetic identities that match no rule, one rule, several rules, a percentage split, and a stale rule set

Needs work: Customer data or shared credentials

Evidence

Ready: identity label, rule version, matching rules, evaluation order, returned variation, cache age, and reason

Needs work: A pass label without observed values

Approval

Ready: Product and release owners choose targeting intent, exposure, kill switch, and production timing.

Needs work: The contractor changes production alone

Reproduce the feature flag precedence failure safely

Two targeting rules match the same identity and different services evaluate them in a different order. Start with one case that can fail without touching customer records or production infrastructure. Write the expected result, stop condition, and reviewer beside the fixture.

Use synthetic identities that match no rule, one rule, several rules, a percentage split, and a stale rule set. Label every identity and artifact so a reviewer can distinguish test evidence from a live record.

Leave enough evidence for a second person , September 11 review

Capture identity label, rule version, matching rules, evaluation order, returned variation, cache age, and reason. Include the source revision and test time. If behavior changes between runs, retain both results.

Product and release owners choose targeting intent, exposure, kill switch, and production timing. A remote programmer can prepare the fixture, run bounded checks, and explain the result. New policy, privileged access, and release approval stay with the named owner.

Put the operating agreement in writing.

Assignment

Test one bounded feature flag precedence case against written acceptance rules.

Fixture

synthetic identities that match no rule, one rule, several rules, a percentage split, and a stale rule set

Evidence

Record identity label, rule version, matching rules, evaluation order, returned variation, cache age, and reason.

Owner decision

Product and release owners choose targeting intent, exposure, kill switch, and production timing.

Frequently asked questions

What should the first run prove?

It should show the expected feature flag precedence result for one controlled fixture.

What belongs in the handoff?

Include identity label, rule version, matching rules, evaluation order, returned variation, cache age, and reason, the source revision, and any untested condition.

Who approves a production change?

Product and release owners choose targeting intent, exposure, kill switch, and production timing.

Sources

  1. NIST Secure Software Development Framework

    NIST guidance for managing software development risk.

  2. GitHub pull request reviews

    First-party documentation for review and approval.

Build a programming lane your technical owner can review.

Bring the stack, backlog, acceptance checks, and release boundaries. We will help scope the role around the work.

Discuss your programming needs