Polish spend should match the cost of the current question, not completeness

What it is

Polish is not a virtue of quality; it is spend that should match the cost of the current question. A cheap question—can the structure be found—gets cheap material. An expensive question—do touch targets hit at live density, does the brand drop trust at a critical conversion—earns pixel-level spend. Chasing “complete” spends money on unasked attributes, delays evidence, and manufactures completeness illusions and wrong-dimension feedback. Matching means every extra hour must be able to change the judgment on that question; otherwise it is over-polish.

Why it happens

Completeness has its own organizational gravity: component libraries, demo habits, fear of looking unprofessional. The gravity points at display, not at the question. Cost is asymmetric—pushing visuals from wireframe to ship-ready is often dearer than another structural round, and changes the structural judgment less. Hours spent polishing are hours not spent enlarging the sample, filling exception branches, or going to the field. Proportionality treats polish as resource allocation inside the experimental design: refine the independent variable where it must be fine, keep the rest disposable. Incompleteness here is a chosen resolution, not laziness.

Studying it

For each question, estimate two costs: the cost of answering wrong (harm after ship) and the cost of polish (hours, tools, locked decisions). Raise fidelity only when the former clearly exceeds the latter. Log where actual hours landed by dimension, and whether comments landed on the same dimension—hours and evidence out of alignment is disproportion. Compare one “complete” round with two cheap rounds plus one targeted polish on problem detection. A budget retro should be able to point at polish that changed no judgment.

Where it stops holding

In some rooms completeness is politeness or a contracted deliverable; rewrite the outward material and keep gaps inward. Accessibility and safety walkthroughs have a statutory resolution that cannot be lowered because “the question is still early.” A one-shot executive review may have no next round, forcing more reality into this artifact; still name the dimensions you made real. Open-source or hiring demos use completeness as a signal; that is communication, not evaluation—do not mix the two kits.

Applying it

• Cap the polish budget in the test plan, and name which question’s resolution that budget buys.
• Refuse “make it complete and then test” when there is no question list.
• At the end of each round ask: which judgment did the extra polish hours change? If no one can answer, cut polish next round.
• Publish the unpolished list to the team so completeness-impulse cannot quietly fill it.

Related

• Same group: Q5.10.1 Mismatched visual and interaction fidelity pulls feedback onto the wrong dimension · Q5.10.2 Stakeholders easily read prototype finish as engineering finish · Q5.10.3 Participants fill in missing pieces and hide genuine confusion
• Adjacent: Q5.01 Fidelity levels · Q5.07 How prototypes mislead
• Search terms: question-costed polish · proportional fidelity · over-polishing cost