[daniele21]

MEASURE — stronger visual hierarchy and semantic palette

• Finding first: one large, high-contrast source-derived finding with a clear evidence label, readable supporting detail, and light/dark-aware semantic foregrounds/backgrounds.
• Actual results next: measured KPI cards with dominant values. Overall quality uses primary brand blue; best local quality and fastest local p50 use teal. The old 'Balance Point' copy has been corrected to 'Best Local Quality' to reflect the existing calculation; published numbers remain unchanged.
• Scope third: models, cases, datasets and runtimes are demoted to neutral contextual cards, after the winners.
• Model inspection fourth: teal indicates Local identity, violet API identity, blue indicates the selected comparison model. Runtime colors do not indicate quality ratings; model roles are still derived at runtime from published evidence.
• Keeps native horizontal rails, real source links, methodology disclosures, benchmark-specific availability states and typography system. Explicit [hidden] guards prevent MCB-only finding/scope appearing for JEV, RedactGuard or unavailable snapshots.

Verification

• ✅ Technical validation passed: Astro build, contract tests and smoke.
• ✅ Homepage visual review passed: viewport 320/390/768/1440 in light/dark, evidence hierarchy, runtime semantic colors, text prominence, scrolling and no page overflow.
• ⏳ The same visual workflow is separately finishing project routes outside this PR scope.

This PR targets dev and will not change main.