What changed
The rebuild changed the public site from a system-map-heavy homepage into an evidence-first product surface. The homepage now explains the failure JudgmentKit prevents, shows the repair path, and routes visitors into value examples, replayable examples, eval evidence, docs, install, and design-system review.
New homepage structure
The homepage uses product-language sections for failure recognition, proof paths, and adoption paths instead of leading with protocol or tool detail.
Disclosure boundary
Raw setup and tool language stays out of the homepage main content. Diagnostic detail moves to docs, evals, install, and this rebuild log.
Proof route
The evals section now includes this log so the rebuild can be inspected as an artifact instead of inferred from chat history.
Dogfood path
The rebuild used JudgmentKit as the planning and review gate before accepting the implementation. The useful signal was not visual taste; it was whether the activity, surface type, disclosure policy, handoff, and implementation evidence were ready.
- Activity model reviewThe brief was reviewed as a public product-site rebuild for AI-agent users and evaluators.
- Candidate repairAn early activity-model candidate exposed raw implementation vocabulary. It was revised before it was trusted.
- Surface selectionThe homepage was treated as a marketing surface with proof and adoption paths, not a setup/debug tool.
- Workflow reviewThe accepted workflow made value, proof, docs, design-system review, examples, evals, install, and MCP setup separate surfaces.
- Implementation contractThe generator stayed static and deterministic, with source-controlled routes, semantic HTML, responsive behavior, and explicit tests.
- Implementation reviewThe final implementation evidence passed after the review evidence was cleaned up and resubmitted.
Design-system evidence
The strongest evidence is in the build and test contract: the same static site generator builds the product pages and the JudgmentKit design-system pages, then exports the manifest, token adapter, component contracts, pattern contracts, specimens, provenance, accessibility policy, and icon scenarios.
- Design-system source
- judgmentkit.ai-native-default.contract-v1
- Token roles
- 9
- Component contracts
- 17
- Surface patterns
- 8
- Icon catalog
- 1737 Lucide icons
- Renderer
- judgmentkit-static-specimens
This page should not be read as a claim that every visual rule on the homepage is mechanically generated from exported token JSON. The defensible claim is narrower: the rebuild is in the same source-controlled static generator, routes users into the JudgmentKit design-system surface, emits the design-system assets in the same build, and has tests that verify those assets, contracts, specimens, and provenance.
Source and tests
The rebuild is inspectable in source and in deterministic checks. These are the files and commands that prove what changed.
Changed files:
site/build-site.mjs
tests/site.test.mjs
Verification:
npm run site:build
node tests/site.test.mjs
npm test
Playwright desktop and mobile review
JudgmentKit review_ui_implementation_candidate: passedHomepage rebuild checks
Tests assert the new headline, proof paths, repair preview, failure grid, adoption paths, and absence of raw setup terms in homepage main content.
Design-system checks
Tests assert the manifest, token adapter, component contracts, pattern contracts, specimens, provenance hashes, accessibility policy, and Lucide icon catalog.
Browser checks
Desktop and mobile review checked no horizontal overflow, visible next-section hint, working menus, and contrast ratios.
Review notes
If you are reviewing whether this is a real rebuild, start with the source diff and tests. If you are reviewing whether it uses the JudgmentKit design system, start with the design-system route, JSON exports, specimen provenance, and the tests that hash those outputs.
The remaining judgment call is product-level: whether the public story should expose more of this evidence earlier, or keep the homepage focused on the offer and leave this page in evals.