Add Simple Icons-style brand-logo corpus flow

[swernerx]

Goal

Create a brand-logo corpus flow, similar in spirit to the Lucide corpus, using a repository such as Simple Icons as the source of SVG brand marks.

Why this matters

Brand logos are one of the most familiar raster-to-vector use cases. A user can immediately understand the quality bar: does the Coca-Cola mark still feel like the Coca-Cola mark, does a bank logo keep its sharp geometry, does an industrial wordmark preserve its negative space and proportions?

This would also give Morphēa a strong comparison story against Illustrator Image Trace, Vectorizer.AI, Potrace, and similar tools: the question is not only "does it look close?" but "does it come back as editable structure?"

Candidate source

Simple Icons looks like the best first candidate:

• GitHub: https://github.com/simple-icons/simple-icons
• Website: https://simpleicons.org/
• npm package: simple-icons
• Provides monochrome SVG paths and brand color metadata.
• The repository is CC0-1.0, but brand/trademark usage still needs care.
• Simple Icons explicitly asks users to read its legal disclaimer and check each icon's license / brand guidelines where available.

Proposed flow

• Add a brand-corpus command or config-driven adapter that reads a pinned Simple Icons package/version.
• Select a small initial subset, for example 12-24 logos across different geometry types:
  • wordmark-like curves;
  • simple geometric bank/finance marks;
  • industrial marks with negative space;
  • multi-part symbols;
  • icons with very thin strokes or tight counters.
• Render source SVGs to PNG fixtures locally, similar to the Lucide corpus flow.
• Run Morphēa back from PNG to SVG/manifest.
• Compare recovered structure against source-SVG-derived targets: shape class, node/parameter budget, visual error, negative-space handling, and editability proxies.

Acceptance criteria

• No third-party brand SVGs or rendered brand PNGs are committed unless the license/trademark position is explicitly reviewed.
• The corpus pins the upstream package version so results are reproducible.
• Reports make trademark-sensitive output easy to keep local-only by default.
• The first smoke uses a small, safe subset and writes artifacts to /tmp or a configured output directory.
• The evaluation distinguishes between visual similarity and editable structure.
• Documentation explains that trademarks remain owned by their respective brands and that corpus usage is for local research/benchmarking.

Notes

Start with Simple Icons because it is consistent and scriptable. Multi-color brand logos or wordmarks from other sources can come later, but licensing and publication constraints should be reviewed before they become public examples.