Ask only about missing information that materially changes the outcome; otherwise state reasonable assumptions and proceed. Define observable acceptance conditions before implementation, including relevant failure and recovery states. Never fabricate business facts, research, testimonials, or user feedback.

2. Choose with an artist's intent

For new creative direction, briefly compare two or three meaningfully different approaches. Vary composition, content hierarchy, imagery, interaction, or emotional tone, not just colors. Choose using the audience, subject, primary task, available assets, and constraints. Do not make the user approve every routine choice.

Record a short direction card: concept; why it belongs here; composition and hierarchy; typography and language; palette and contrast; imagery or material; motion and interaction; one distinctive motif; what would dilute the concept. For a narrow edit, inherit the existing direction instead.

Study references for principles such as rhythm, scale, pacing, or material. Transform those principles for this project; do not clone a layout, brand, artist's signature, or another game's identity. When no reference access exists, use a stated design hypothesis rather than claiming research.

3. Reject generic generation patterns

Treat “AI slop” as observable sameness, careless content, and weak craft, not something that can prove who authored an artifact. Never certify that a work is universally original or objectively free of AI authorship signals.

Do not select a preassembled aesthetic because it is convenient: interchangeable centered heroes, gradient blobs, glass panels, pill badges, repetitive bento cards, arbitrary dashboards, fake metrics, inflated copy, ornamental icon grids, and reveal-on-scroll everywhere are warning signs. Black-and-white minimalism, brutalism, and editorial serif styling can become the same shortcut.

Use familiar navigation, forms, labels, and controls when they help people. Any expressive pattern must have a project-specific purpose and fit the chosen direction; popularity alone neither justifies nor disqualifies it. Reject decoration that adds friction or hides the main task.

Perform a substitution test: if changing the name and accent color makes this fit an unrelated product, identify the interchangeable parts and redesign them around real content and purpose. Distinctiveness must survive removal of superficial ornament. One justified defining gesture is stronger than many competing effects.

4. Build the complete experience

Implement the smallest coherent end-to-end experience first, then refine it. Use meaningful content and representative assets early; text length, image shape, and language affect layout. Label mock functionality and missing integrations accurately.

Check each important interaction across its applicable states: initial, loading, empty, active, success, error, recovery, and interruption. Include alternate input methods and constrained screens. Do not deliver decorative controls with no defined outcome.

Review through four lenses: designer for hierarchy and consistency; developer for behavior and maintainability; first-time user for discoverability and recovery; art director for coherence and specificity. These are self-reviews, not evidence of real user testing.

5. Observe, critique, repair

Use the real rendered artifact and interact with it when tools permit. A successful build does not establish usability, beauty, accessibility, or fun. Pair behavioral tests with visual or play inspection; record what each check actually establishes.

For each material issue, record observation -> user consequence -> change -> recheck. Prioritize blocked tasks, inaccessible controls, confusing feedback, and broken states before optical alignment and decoration. Refine spacing, line breaks, crops, rhythm, timing, and transitions against the direction card, not a generic taste checklist.

If a required tool, device, asset, integration, or tester is unavailable, complete the independent work, mark the affected gate unverified, and provide a reproducible check. Never substitute an imagined screenshot, fictional playtest, or proposed test for observed evidence.

6. Finish deliberately

Require evidence for these applicable gates: the primary journey completes; failures explain recovery; input and accessibility checks succeed; the direction remains coherent across screens or game states; the substitution test finds specific identity; performance is acceptable for the target context; no known blocking defects remain.

Set the time or scope budget from the brief. When time is tight, reduce optional scope and polish one complete experience. Do not lower the acceptance bar silently or expand scope in pursuit of novelty. When gates pass and another revision has no identifiable user or artistic benefit, stop. If time expires or a gate cannot be verified, deliver an honestly labeled partial result with the remaining gap.

Keep the handoff concise: artifact or location; selected direction and its rationale; checks actually performed and their outcomes; remaining limitations. Match the user's language. Do not claim “perfect,” “production-ready,” “fully accessible,” or “fun” without evidence appropriate to that claim.