Chapter 5
Prototyping to resolve uncertainty
Choose prototype fidelity, evaluate a meaningful task, and use the evidence to decide what to build next.
The team can demonstrate the new eligibility message in a few seconds. That proves the prototype can display it. It leaves open whether administrators understand the message, choose the right action, and recover when the situation changes.
Choose the question before choosing the prototype’s fidelity. For this case, the immediate question is whether visible eligibility information helps administrators decide what to do before and after an assignment attempt.
Match the prototype to the uncertainty
A static layout can support a critique of hierarchy and language. A clickable prototype can help explore sequence and navigation. A coded prototype can represent changing state, keyboard behavior, and a delayed or lost response. A technical experiment can establish whether the service exposes the facts the interface needs.
You may need more than one artifact. The Casework team uses a small coded flow for recovery behavior and a service conversation for eligibility semantics. The flow does not need a full search system or a production authentication integration to test its current question.
Keep an assumptions list beside the prototype. Examples include “case status is supplied by the service,” “the administrator can read the current assignment,” and “the chosen specialist team remains valid.” An assumption demonstrated with invented data has not become a verified service capability.
Write a task that leaves room for discovery
“Click Assign and check the success message” tells participants which path to follow. A more useful task is: “You are clearing the morning queue. Route this case to the appropriate team, and tell us when you are confident it is ready for them.”
Provide enough context to make the choice meaningful. Include a case that cannot be assigned, a straightforward case, and a case whose status changes during the attempt. For a lost response, ask the participant what they believe happened before directing them anywhere else.
Agree on the evaluation with a researcher where possible. Decide what to observe, what questions to ask afterward, and what limitations to record. Test the task wording in a pilot so an ambiguous brief does not become a false interface finding.
Capture observations and decisions separately
Here is a fictional example for discussion; no sessions produced these observations:
| Supplied observation | Interpretation to investigate | Proposed response |
|---|---|---|
| An administrator reads the closed-case explanation and continues to another case | The explanation may support the intended decision | Keep it and evaluate other blocked states |
| After a lost response, someone submits again because the selected team still looks editable | The interface may imply that no assignment occurred | Explain uncertainty and require reconciliation before another write |
| A participant asks whether the visible eligibility is guaranteed | The interface may overstate the freshness of its information | Clarify that the service checks again during submission |
The first observation concerns one case in a limited setting. It does not establish a success rate. The second identifies a consequential risk worth addressing before release, even without an estimate of its prevalence.
Do not combine opinions, observations, and researcher interpretations into an undifferentiated list of findings. Preserve enough context that another person can assess the claim.
Test the difficult path again
Revise the uncertain-outcome state. Explain that the assignment may have saved, prevent another write until the latest state is checked, and make that check available. Evaluate whether the person understands both the uncertainty and the recovery action.
A revised message is a design hypothesis. If the first evaluation exposed a significant problem, inspect the changed behavior with another appropriate task instead of assuming that writing the fix resolved it.
Also ask what the prototype cannot tell you. A local simulation cannot establish production response times, backend authorization, realistic support burden, or whether people adopt the workflow during actual operations.
Make the next decision explicit
Your evaluation ends with a recommendation: proceed, revise, narrow, stop, or investigate further. Link it to the evidence and remaining uncertainty. For the worked case, a reasonable recommendation is to proceed with a single-case production slice after checking that the service can return a known outcome and expose the latest recorded assignment.
The product lead still needs to accept scope, and engineering needs to validate the contract. A successful prototype evaluation does not create either agreement automatically.
Practice: run a small evaluation
Write one decision, the uncertainty behind it, and the smallest prototype that can address it. Define realistic tasks and the observations that would change your recommendation. Use appropriate participants if available; otherwise conduct a clearly labeled peer exercise and identify which domain questions remain unanswered.
Record what happened, your interpretation, a proposed change, and a follow-up check. Produce a short recommendation with an assumptions list. Keep actual observations separate from the fictional evidence supplied in this chapter.
Review your work
A reviewer should understand what the evaluation established and what it could not establish. The recommendation should follow from the evidence rather than the effort invested in the prototype. At least one observation should be capable of changing your direction.
The next chapter turns those assumptions into an implementation contract and a working flow.
Found an error? Report it.