Skip to main content
UX Design Engineering
Settings

Settings

Theme
Text size
100%
Typeface

Chapter 7

Establishing quality and releasing safely

Gather release evidence around consequential behavior, define support ownership, and plan a staged rollout with clear stop conditions.

The assignment flow works on a developer’s machine. It saves the chosen team and displays confirmation. Before release, someone asks what happens if a keyboard user receives a stale-case rejection or the service records the assignment without returning a response.

“The happy path works” cannot answer those questions. Release readiness needs evidence about the behavior that matters and a plan for responding when production differs from the example.

Turn acceptance criteria into checks

Start with the behavior model and scope decision. Identify the failures that could mislead someone, lose their work, route a case incorrectly, or prevent access. Choose a check that can expose each one.

Consequential behaviorAppropriate evidence
The server rejects unauthorized writesIntegration evidence against the real service, including direct requests
A stale case does not receive a silent overwriteService contract check and an interface scenario with a changed version
A lost response does not trigger an automatic duplicate writeControlled response-loss scenario and request inspection
Keyboard users can complete and recoverManual keyboard evaluation of the complete task
Status and error information are available to assistive technologyScreen reader evaluation in the supported environment
Large or slow responses do not leave the interface indefinitely pendingRealistic network and data scenarios with defined timeout and recovery behavior

Check the parts at the level where their guarantees live. A unit check of browser code cannot prove server authorization. A successful HTTP response cannot prove that the user understood the assignment outcome.

Use automated and manual evidence together

Automate stable, consequential behavior where the project’s tools support it. Useful examples include rejecting invalid service inputs, retaining input after a known rejection, and preventing a second write while an operation is unresolved. Avoid checks that merely repeat component internals without protecting a user outcome.

Inspect the whole flow manually. Complete it with a keyboard, enlarge text, and evaluate it with supported assistive technology. Check that the live region communicates an understandable outcome, that essential text remains visible, and that the next action can be found after a failure.

For the local example, distinguish observed behavior from intended behavior in your notes. The simulation can demonstrate a client decision; it cannot validate production authentication, actual network conditions, or an organization’s accessibility conformance.

Record limitations with owners

A useful release note might say: “Stale-state recovery passed in the staging scenario. Screen reader review found that the updated explanation was announced twice. The feature engineer will remove the duplicate announcement and the specialist will repeat that path before the pilot.”

That note identifies evidence, a remaining issue, and a next check. “Accessibility done” conceals those details.

Use the same discipline for performance. Record the tested device, data size, network condition, and task. Choose budgets appropriate to the product rather than declaring a universal number. If checking the latest case can stall, agree on a timeout and explain how uncertainty remains visible after it expires. The tiny local simulator does not implement a real network timeout policy.

Plan the release as a reversible step

For the fictional Casework pilot, begin with an agreed operations cohort. Confirm support coverage and the ability to disable the new interface. Preserve access to necessary existing workflows while evaluating the change. Check that disabling a feature does not imply undoing assignments already recorded by the service.

Name who monitors errors and support questions, who decides whether to pause, and how people reach them. Verify the telemetry before depending on it. A missing event should not become evidence that no failure occurred.

Define stop conditions in terms of consequences: an authorization defect, incorrect routing, a critical accessibility barrier, or an unexplained increase in unresolved outcomes. Agree on how to investigate and respond. Expansion requires another decision; it should not occur merely because the pilot period ended.

Practice: hold a release review

Choose several consequential scenarios from your behavior model. Run the relevant checks and record environment, observation, result, and limitation. Where you lack access, write the proposed check and its owner without marking it complete.

Prepare a short release plan containing the cohort, quality evidence, known issues, instrumentation, support contact, stop conditions, and recovery approach. Ask a peer to introduce a failure while reviewing it. Explain who acts and what can actually be reversed.

Review your work

A reviewer should be able to trace each important release claim to an observation or an appropriately scoped automated result. Unperformed checks remain visible. The release plan should identify who supports the workflow after launch and which circumstances prevent expansion.

For component-level techniques, consult testing accessible components. The next chapter examines what the pilot can tell us about product effectiveness.

Found an error? Report it.