Prototype fidelity, test planning and experimental learning

Fidelity has several dimensions

A prototype is a representation built to learn about selected aspects of a proposed solution. Fidelity describes resemblance in relevant respects. Appearance, dimensions, interaction, materials, behaviour and integration can differ independently. Calling one prototype “high fidelity” without naming the property can therefore hide the very limitation that matters.

An interactive paper interface can test whether a person understands a sequence while a facilitator changes the cards. It cannot establish the reliability of the real network service it depicts. A full-size cardboard tray can preserve reach distances and obstructed access while deliberately omitting final material behaviour. State both the preserved feature and the excluded inference.

Build the representation that can expose the uncertainty

For a label-comprehension question, prepare alternative instructions and a consistent task. For an access question, reproduce the relevant dimensions, surrounding obstacles and posture. For a local joining question, a representative material/process coupon may be useful, but a successful coupon does not establish the behaviour of the complete assembly.

The matrix is a proposed plan, not data. It connects each object to what it can reveal and what it cannot establish. One prototype may answer more than one question, but adding details has value only when those details improve the needed evidence. A test that cannot distinguish the competing explanations may need a different representation rather than a more polished one.

Write down what would change the design decision

Our original tray exercise compares two latch positions. Specify the task, relevant users, surrounding space, starting condition, assistance rule and observation method. Record successful unaided access, time if relevant, errors and comments. Decide in advance how the evidence will inform the next design step. This is a formative comparison, not a statistical claim about a whole population.

Control obvious differences that would prevent interpretation. If one arrangement is always tried after a demonstration and the other is not, a difference may reflect learning. If the intended task involves a loaded tray, an empty cardboard trial cannot establish that loaded performance. Preserve unexpected observations; they can expose a missing requirement or the need for a different test.

Turn an access question into a useful trial

Complete one part of the tray exercise before comparing alternatives. The question is: can intended users identify and touch the proposed release tab while the empty tray remains in its shelf bay? Build a full-size mockup that preserves the bay side walls, front lip, tab location and hand space. Fix the parts so they do not bend away when touched. Omit the working catch: this trial concerns locating and reaching the tab, not the force or motion needed to release it. A reduced model or a loose cardboard wall would change the geometry being investigated.

Invite people who would use the tray, including the postures and access needs relevant to the brief. Explain the mockup and ask each person, from the same starting arrangement, “Show which part you would use to release the tray, then try to touch it with the tray still in place.” Do not point out the tab during this first attempt. Record the first part selected, whether contact with the intended tab is possible, any collision with the lip or walls, and what the person says. Record requests for help separately. These observations connect the task to the question; a general preference rating would not reveal the same obstruction.

Plan a separate diagnostic step after recording that unaided attempt. If someone selects another part or cannot find the tab, point out the intended tab and ask whether they can now touch it. Mark that observation as assisted. It cannot turn an unsuccessful first attempt into an unaided success. It can help distinguish not knowing where to act from not being able to reach the place once it is known.

Consider two possible observations, not invented test results. If people correctly identify the tab but their fingers meet the lip before reaching it, revise the access geometry and retest it in the represented bay. A more prominent label alone would leave that obstruction. If people search elsewhere but can reach the tab after it is identified, revise the cue or placement and try the unaided task with people who have not learned the answer. If both occur, both design questions remain. Do not collapse these records into one “passed” count.

This plan produces formative evidence for the people and conditions observed. It does not prove population-wide access, release force, stiffness with a load, durability or performance with the final catch. The next prototype should add the omitted behaviour when the design question moves to it. That change is purposeful fidelity: build the property needed for the next decision rather than adding polish indiscriminately.

Use a scale calculation to reject an unsuitable trial

A 1: 2 linear-scale model of an 80 mm access opening has a 40 mm opening. That is useful for discussing overall arrangement, but it cannot directly test whether a real adult hand fits through the full-size opening. The hand has not been scaled.

For geometrically similar solids, area scales as the square of the linear factor and volume as its cube: here 1/4 and 1/8. Mechanical behaviour is not automatically preserved by those factors. The calculation identifies why a convenient small object should not be treated as a direct physical performance test.

Try the reasoning

1. Can realistic appearance establish realistic behaviour?

   Show the answer and reasoning

   No. The property relevant to the decision must be represented and checked; appearance is only one dimension of fidelity.

2. What can the cardboard access trial usefully show?

   Show the answer and reasoning

   Whether the represented full-size geometry and conditions obstruct the selected task. It cannot establish final-material stiffness or lifetime.

3. Why define the assistance rule before an interface trial?

   Show the answer and reasoning

   A facilitator’s hints can change the task and obscure whether the person understood the proposed interface independently.

4. An exhibit team wants to know whether visitors understand how to choose an audio language before starting a story. It has a sequence of interactive paper screens and a realistic screenshot that cannot respond. Select a first prototype and give a concrete task, observations and next action. Can the same trial establish touchscreen accuracy or audio loading time?

   Show the answer and reasoning

   Use the interactive paper screens to represent the available language choices, wording and resulting confirmation, with a facilitator following the same response rules. A neutral task could be “Start the story in Spanish.” Watch which route visitors choose, whether they understand the confirmation, and where they backtrack or ask for help. If the route is found but the confirmation is misunderstood, revise that message; if the language choice is not found, investigate its location or label. Try the revised version without teaching new participants the answer first. The static screenshot cannot reveal the sequence. Neither paper trial establishes real touch-target accuracy or loading time: those questions need the relevant interaction and system behaviour represented and measured.

What to carry forward

Make the prototype realistic where the question needs it, and interpret observations within that declared scope.