Skip to content

v2.11.0 — distinct directions, proven on real pages

Choose a tag to compare

@Kayforkind Kayforkind released this 06 Sep 09:11
· 30 commits to main since this release
5003bd3

What changed

  • Unique layout grammar per direction. Converging clusters now carry their own compositions: cinematic is a filmstrip + timecode chapter grid, photography is monospace-captioned plates with no overlays, motion draws labeled orbit rings carrying the anchors, dashboard ships a station-index head-sheet, infographic ranks source numbers as scale bars, editorial gets a data sheet, lookbook drops the glyph chips. Mean class-set difference between directions on the same source: 94.6% → 96.5%.
  • Public-source proof lane. examples/public-sources/ rebuilds real U.S. government pages — NPS Yellowstone bulletin → editorial (84%), NASA Artemis II → 3js (100%), NOAA/NWS hurricane outlook → landing (100%). No declared palettes; color derived from language alone. Reproduction guard now covers 14 committed artifacts byte-identically.
  • Claims parity. Six stale alternates rows in the README example table corrected against regenerated Auto reports; docs site gains the public-proof section; intake labels (public-source, auto-routing, token-distinctness) filed with tracking issues #38/#39.

Verification

  • npm test: 12/12 phases green
  • Reproduction guard: 14/14 artifacts byte-identical at pinned seeds
  • Benchmark: 17 tokens × 4 sources, all 100/100 usability, full fidelity
  • npm: reimagine-it@2.11.0 with SLSA provenance; release assets cosign-signed

Full notes: CHANGELOG


🛒 GitHub Actions Marketplace

This release publishes the reimagine-it Design Health action to the GitHub Actions marketplace.

Quick start

- uses: Kayforkind/reimagine-it@v2.11.0
  with:
    path: "**/*.html"

What it checks

Category Checks
Typography Hierarchy levels, text measure, banned default fonts
Palette Off-palette accents, unstyled ::selection, transition:all
Motion Missing prefers-reduced-motion, missing :focus-visible, non-compositor animation
Content Placeholder copy, vibe adjectives, emoji farm
Structure CDN / external font fetch, figure system
Performance Image dimensions, content-visibility

19 deterministic rules. No LLM. No API key.


📊 Measured quality (auto-generated)

Direction Fidelity (title kept) Usability (quality /100) Content art (glyphs/donut/bars/prism)
webpage 18/18 100/100 5.0
landing 18/18 100/100 10.0
dashboard 18/18 100/100 8.8
infographic 18/18 100/100 8.0
cinematic 18/18 100/100 16.0
artistic 18/18 100/100 11.0
photography 18/18 100/100 8.0
svg 18/18 100/100 6.0
3js 18/18 100/100 7.0
simulation 18/18 100/100 9.0
glass 18/18 100/100 10.0
editorial 18/18 100/100 4.0
motion 18/18 100/100 5.0
gradient 18/18 100/100 10.0
showcase 18/18 100/100 10.0
lookbook 18/18 100/100 10.0
particles 18/18 100/100 2.0
Output diversity: 18.6% mean pairwise size difference, 96.5% mean class-set difference between directions on the same source — no two tokens produce the same page.
Coverage: 68 generated pages (17 tokens × 4 sources), all audited structurally. Regenerate with node scripts/benchmark-tokens.js.