SuperAPI AI Directory

prototype-lab

General & other

View repository

Updated 10-16 05:45

About this skill

Prototype Lab

Prototypes answer decisions. They are not low-quality versions of production and not a collection of minor style tweaks.

The entire value of this skill is divergence. Three tints of one idea waste the switcher — the user learns nothing by flipping between them.

Quick reference

Need Read
Product job, hierarchy, restraint, working order references/interface-principles.md
Radius, optical alignment, elevation, materials, gradients, icons references/surfaces-and-depth.md
Whether to animate, easing, duration, interruptibility references/motion-principles.md
Drag, swipe, velocity, boundaries references/gesture-physics.md
Semantics, keyboard, focus, hit areas references/accessibility-contract.md
Evidence boundaries and feel checks references/evidence-and-verification.md

Divergence is not permission to drop the craft bar. Every variant individually meets the non-negotiables — correct easing, sub-300ms UI motion, correct transform-origin, transform/opacity only, real focus states, reduced motion handled. A sloppy variant does not widen the exploration; it loses on execution and teaches nothing about the direction it represents.

Define the decision

Establish:

  • the product question the prototypes must answer;
  • what is fixed — content, brand, layout, technology, accessibility, and performance constraints;
  • what may vary;
  • target device and input method;
  • how often the user will encounter this surface — it bounds how far the boldest variant may go;
  • the product's personality: a crisp daily-use tool earns quieter variants than a playful consumer app;
  • expected fidelity;
  • how many directions to explore — recommend 3, up to 5 when the space is genuinely wide. More than 5 dilutes the comparison;
  • evaluation criteria, and who chooses.

Do not start until the variants can be compared against the same question.

One thing per run. If the description spans several components, narrow it: pick the single highest-leverage piece, say which and why, and offer the rest as follow-up runs.

Create distinct hypotheses

Each direction must differ in its underlying interaction, hierarchy, spatial model, or material logic. State each variant's axis in a phrase before building anything. No two variants may share an axis position.

Weak variation Strong variation
Same component, different accent color Direct manipulation vs. state-triggered transformation
Same motion, small duration change Spatial continuity vs. a deliberate cut
Same layout, different radius Geometry-led morph vs. layered material transition
Same structure, different copy Compact disclosure vs. persistent overview
Restrained functional motion vs. rare expressive choreography

Name each concept by its hypothesis — "Quiet", "Editorial", "Dense", "Direct" — never "Option A/B/C" and never an adjective like "modern".

Sharing the project's tokens is not convergence. Variants should look like they could ship in this product tomorrow.

Write concept cards

Before code:

### Concept name
- Hypothesis:
- Axis it explores:
- User benefit:
- Interaction model:
- Visual and motion model:
- What stays fixed:
- Main risk:
- What the prototype must prove:

Reject redundant concepts before implementation. If two converge while you build them, cut one and say so — two truly distinct directions beat three padded ones.

Build an isolated lab

  • Never touch production code during exploration. Keep prototypes outside production routing and state unless the user asks otherwise.
  • Reuse the product's tokens and primitives when they are part of the question.
  • Use realistic content — actual product-shaped copy, plausible names and numbers, target dimensions. No lorem ipsum, no dead buttons, no "imagine this part".
  • Give every variant the same baseline constraints.
  • Add a switcher with stable concept IDs. Switching is instant — flipping is a high-frequency action within the session, so by the frequency gate the variant swap itself gets no animation.
  • Render one variant at a time, full size, in realistic surrounding context — a toast needs a page behind it, a card needs siblings, a button needs a form. Side-by-side thumbnails distort spacing and scale.
  • Preserve the selected concept in URL state when sharing or repeatable comparison matters.
  • Include reset, pause, replay, and reduced-motion controls when motion is under evaluation.
  • Show key parameters only; do not turn every constant into a control.

Avoid premature abstraction between variants. Duplication is acceptable while the concepts are intentionally diverging; extract shared code after selection.

Verify every concept

For each variant, capture:

  • normal-speed behavior;
  • slow or stepped inspection where motion is under evaluation;
  • keyboard and touch behavior, including visible focus;
  • the reduced-motion version;
  • narrow and wide layout;
  • interrupted and repeated interaction;
  • known performance cost;
  • limitations and unresolved questions.

Flip through every variant yourself before showing the user. Confirm each renders, each interaction responds, and the console is clean.

Do not polish one favorite more than the others before comparison. An unfair comparison answers the wrong question.

Selection gate

Present the comparison and stop. The choice belongs to the user.

Concept Axis When it wins What it costs Accessibility Performance Complexity

Sell each variant honestly — one line on when it is the right choice, one on what it costs. Never pre-pick a favorite inside the table. If the user asks which you would choose, answer with a reason rooted in the product's personality and frequency of use, not aesthetics alone.

Close with where the lab is running and how to flip between variants.

Do not silently promote a prototype into production code.

After selection

  1. record the chosen concept and the rejected alternatives;
  2. define the production acceptance criteria;
  3. identify prototype shortcuts that must be replaced;
  4. move only the selected behavior into the product, following the project's existing conventions;
  5. remove or archive the lab only with user approval.

If the user wants another round instead, keep the harness and diverge around the direction they gravitated toward.

Common mistakes

Mistake Fix
Three variants of one idea State each axis first; no two share a position
Variants named "Option A/B/C" Name by hypothesis
One favorite polished more than the rest Equal effort until the comparison
Lorem ipsum or dead buttons Realistic content, working interactions
Side-by-side thumbnails One variant at a time, full size, in context
Animated variant switching Instant — it is a high-frequency action
Craft bar dropped "because it's a prototype" Every variant meets the non-negotiables
Shared abstraction extracted too early Duplicate while diverging; extract after selection
Prototype promoted without asking The selection gate is required
Lab left in the repo silently Remove or archive only with approval