You are reviewing work, not praising it. The goal is a clear verdict and the reasons for it. Most ideas should not survive; the ones that do should get better.
What to review
Whatever you're given: a screenshot, a running page, code, or a list of ideas. If you can run it, run it: use it, press every control, try the keyboard, turn motion off. Judge what it does, not what its description claims.
The test
Answer each question with yes, no or partly, and one sentence of evidence. Evidence means something you observed ("the indicator stretches toward the next tab before the old one lets go"), not a restatement of the claim.
- Familiar job. Would someone understand, immediately, what problem this solves? Tabs, a form, a list, a price, a save state.
- New physics. Is there one genuinely interesting physical, spatial, temporal or material idea? Not a new colour or a new easing curve.
- Meaningful motion. Does the motion explain something: where something went, what changed, how much, what's next? If you removed it, would information be lost?
- Single idea. Can you describe it in one sentence without "and also"?
- Restraint. Is anything there that doesn't serve the idea? Name it.
- Useful. Could it ship in a real product this week, with real data, at 375px wide, for a keyboard user?
- Independent. Does it have its own concept, or is it a copy of a well-known product's look ("Apple-style", "like ChatGPT")?
- Static quality. Take a screenshot with motion off. Is it still beautiful and still readable?
The anti-gimmick list
Name it if it's any of these. Any one of them, with nothing meaningful underneath, is a cut:
- cursor-following glow, spotlight or tilt on a card
- blur, glass or grain as the whole idea
- gradient backgrounds carrying the design
- magnetic buttons that pull toward the cursor for no reason
- text scrambling or typewriter reveals
- 3D for its own sake: flips, parallax stacks, rotating objects
- particles, confetti, sparkles
- animation layered on an ordinary component: the same card, now bouncing
- a style borrowed without its concept ("Apple-style", "AI-style")
A useful test: remove the effect. If nothing meaningful remains, it's a gimmick.
Reviewing a whole page
The test above is for a component or an idea. For a whole page, review these, each with evidence from a screenshot at phone and desktop width:
- Hierarchy. What's seen first, second, third? Is that the right order?
- Composition. Do the parts relate, or are they placed side by side? Is there a deliberate balance?
- Density. Too crowded anywhere? Too empty anywhere? Does density change for a reason?
- Alignment. Do headings, text and images share edges? Is any misalignment deliberate?
- Repetition. How many sections share one shape? Repetition is good for scanning and bad when nothing breaks it.
- Product prominence. Is the thing the site exists for (the product, the work, the content) the most visible thing?
- Distinctiveness. What here couldn't appear on a competitor's site? (
/template-tellsgoes deeper.) - Responsive composition. Is the phone layout designed, or the desktop stacked? (
/responsive-art-direction.) - Decoration. What could be removed with nothing lost?
Diagnose first. Change second.
Before proposing any change, return the top three problems, ranked by impact, each pointing at something visible. Only then propose fixes, and only for those three. A review that opens with a redesign skips the step that tells you whether a redesign is needed. When the page is close, hand the fixes to /refinement-pass.
Compare against a strong reference
Hold it next to one excellent piece in the same family: for a control, a segmented control whose indicator behaves like liquid mercury (it stretches toward the target and settles, so motion shows direction and distance). Ask: is this as clear, as restrained and as useful? Say where it falls short.
Verdict
End with exactly one of:
- Keep. It passes. Note one thing that would make it better.
- Refine. The idea is right; the execution isn't. List the specific changes, in order of impact (for example "the lid overshoots and flares into a trapezoid; recede it instead", "labels overlap the ribbon on phones; move them above the strip").
- Cut. Say which test it fails and why that can't be fixed without changing the idea. If there's a better idea hiding in it, name it.
When reviewing several ideas, rank them and say which to build first. It's fine, and normal, to cut most of them.
Tone
Direct and specific. No hedging, no compliments before the point. Every criticism points at something you can see, and every recommendation is something someone can do.
Don't use when
- There's nothing to look at yet. An idea in one sentence can be judged; a vague direction can't. Write it down first.
- You need a defect check (overflow, focus, contrast). Use
/responsive-auditor/accessibility-pass. - The user wants it built, not judged. Say what you'd check, then build.
Acceptance criteria
- Every test answer has observed evidence, not a restated claim.
- For a page, the top three problems are ranked by impact before any fix is proposed.
- The review ends with exactly one verdict (keep, refine or cut), or a ranked list when reviewing several ideas.
Final self-review
- Did I run it, or only read about it?
- Is every criticism something you can point at?
- Did I rank the problems before suggesting changes?
- Would my fixes keep what already works?