Corsac
Use Corsac
All evals
P

Eval directory

Evals for Polymet

Eval coverage for Polymet, mapped from its public product surface.

About Polymet

Polymet is an AI product designer that generates pages, components, and prototypes and returns production-ready frontend code. It supports importing your own design system, Figma import/export, and both visual and code editing. It integrates with dev workflows (GitHub, npm, Storybook) and offers team collaboration with shareable live demos.

Industry

AI product design tool (design-to-code)

Website

polymet.ai

Use the eval library for Polymet

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

Generate your own →

Related

All evals →

More AI eval libraries

Coverage map

What would you measure for Polymet?

6 scoring areas · 22 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

  • Adversarial robustness
  • Workflow quality
  • Safety gates
  • Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

01

Design Generation from Intent

Turning a prompt, an image, or a product brief into pages, components, and clickable prototypes — the primary entry point advertised on the landing page ('Design a product', 'Design a component', 'Design from image').

“You can design your pages, components and prototypes with Polymet and get production-ready frontend code.” polymet.ai

Mapped capabilities

4 capabilities

  • Full-page generation from a product brief

    Layout, hierarchy, and states of a complete page generated from a natural-language description.

  • Single-component generation

    Generating one reusable component with its variants and props rather than a page-scoped one-off.

  • Design from image

    Reconstructing layout, spacing, and component structure from an uploaded screenshot or mock.

  • Prototype flows

    Multi-screen prototypes with navigation between generated pages.

02

Design System Fidelity

Whether an imported design system is actually honored — reusing existing components and tokens, iterating existing ones rather than forking them, and extending the system consistently when something genuinely new is needed.

“Bring your own design system, create pages and new components and iterate existing ones.” polymet.ai

Mapped capabilities

4 capabilities

  • Reuse of imported components

    Generated output composes existing library components instead of re-implementing lookalikes.

  • Token and scale adherence

    Colors, type, and spacing resolve to design-system tokens rather than arbitrary literals.

  • Iterating an existing component

    Modifying a component in place while preserving its API and existing variants.

  • Consistent extension for new components

    New components introduced in the system's naming, variant, and composition conventions.

Illustrative example

Input
Imported design system defines a Button with variants primary, secondary, and ghost, plus a 4px spacing scale. Prompt: "Add a pricing page with a call-to-action button."
Expected behavior
The generated page renders the imported Button in one of its existing variants and uses spacing values from the 4px scale. It does not declare a new inline button style or invent arbitrary padding values.

03

Code Output and Dev Workflow

The 'production-ready frontend code' claim and the integrations that carry it into a real repo: GitHub, public and private npm packages, and Storybook.

“Works with Github, public & private npm packages, and storybook.” polymet.ai

Mapped capabilities

4 capabilities

  • Code correctness and buildability

    Emitted code typechecks and builds without hand-editing.

  • GitHub round-trip

    Branching, committing, and opening changes against a connected repository with a contained diff.

  • npm package usage

    Correct import and usage of public and private packages already in the project.

  • Storybook stories

    Generated components ship with stories that render the documented variants.

Illustrative example

Input
Connected GitHub repo with a React + TypeScript + Storybook setup. Prompt: "Generate a Card component with a title, body, and optional footer, then open a pull request on a new branch."
Expected behavior
Polymet emits a typed component file and a matching Storybook story, commits them to a new branch, and opens a pull request against the default branch. Unrelated files in the repository are left untouched.

04

Figma Interchange

Import and export fidelity between Figma and Polymet, including whether structure survives a round trip rather than flattening into unstructured layers or markup.

“Import and Export Figma files, level up your design experience.” polymet.ai

Mapped capabilities

3 capabilities

  • Figma import fidelity

    Frames, auto-layout, and text styles map to structured components rather than flat output.

  • Figma export fidelity

    Generated designs export as editable, well-named Figma layers.

  • Round-trip structure preservation

    Component identity and naming survive import → edit → export.

05

Editing Control (Visual and Code)

The advertised 'full control' story: tweaking layout, spacing, and colors visually, editing the underlying code directly, and keeping the two representations in agreement across iterations.

“Tweak layouts, spacing, colors, and more - visually.” polymet.ai

Mapped capabilities

4 capabilities

  • Visual edits to layout and spacing

    Direct manipulation produces the intended change without disturbing sibling elements.

  • Direct code editing

    Hand-edited code is preserved rather than overwritten by later generation.

  • Visual/code synchronization

    A change made in one editor is reflected faithfully in the other.

  • Iterative refinement

    Follow-up instructions scope to the targeted element instead of regenerating the screen.

06

Collaboration and Sharing

Team review and stakeholder sharing: real-time collaboration on a design and live demo links opened outside the editing team. Security and compliance appear on the site as an enterprise claim; the public evidence supports only shallow coverage here.

“Design with your team, review fast, and refine in real time.” polymet.ai

Mapped capabilities

3 capabilities

  • Multi-user collaboration

    Concurrent edits and review on a shared design.

  • Shareable live demos

    A share link renders the current prototype for a non-editing stakeholder.

  • Access scope of shared links

    Whether a demo link exposes only the intended design and nothing else in the workspace.

Coverage is mapped from Polymet's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Polymet test?+

The coverage map is generated from Polymet's own public product surface (AI product design tool (design-to-code)): 6 scoring areas — Design Generation from Intent, Design System Fidelity, and Code Output and Dev Workflow, and more — spanning 22 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Polymet evals scored?+

Every case generated for Polymet — across Design Generation from Intent and Design System Fidelity and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Polymet library include?+

The full Polymet library is built on request. The coverage map spans 6 areas and 22 capabilities (for example, Full-page generation from a product brief and Single-component generation under Design Generation from Intent); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Polymet or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Polymet areas and set them up in a Corsac workspace, where you can run every test case against Polymet or your own agent with your own data.