Improve homepage onboarding and document answerability results

[RodCor]

Make the homepage explain Kimetsu through a concrete cross-session memory example and give visitors a direct path to installing it. Publish the unreleased answerability study with its configuration, gains, costs, and unresolved cases.

Website experience

• Replace the metric-heavy hero and comparison tabs with a memory workflow, a concise explanation, and clear links to installation and evidence.
• Add an agent selector for Claude Code, Codex, Cursor, Pi, and OpenClaw, with copyable setup commands and accessible success/failure feedback.
• Make homepage content server-rendered and visible without JavaScript. Remove observer-based reveals and their animation code; keep client state limited to setup interactions within the homepage content.
• Add practical FAQs, more useful navigation, improved metadata, visible keyboard focus, a first-tab skip link, and reduced-motion support.
• Fix light-theme text contrast and the navigation's decorative repository icon. Keep mobile content within the viewport and provide native keyboard controls.
• Replace absolute model-free/zero-token language with explicit distinctions between local compute, optional model calls, and agent context costs. Existing historical benchmark reports remain available in the docs.

Answerability documentation

The new results page reports 15/18 → 3/18 unwanted injections with 24/27 positive hits retained, exact metadata at 36/45 per repeat, p95 latency of 376.6 → 386.6 ms, and mean response bytes of 613.5 → 651.2. It labels the guard unreleased, opt-in, and disabled by default.

Update benchmark navigation, configuration guidance, and the changelog. Link the commit-pinned public audit and explain synthetic-fixture limits, compound-query gaps, storage work, and context-token use. These measurements do not replace historical BrainBench scores or measure generated-answer accuracy.

Validation

• npm run check: Biome, type checks, all 29 tests, Next.js production build, and static-output validation passed.
• Chromium browser checks: host switching, exact clipboard contents (normalized Windows line endings), denied-clipboard feedback, native FAQ expansion, internal homepage link responses, and keyboard skip navigation.
• No horizontal overflow at 320, 375, 768, 1024, or 1440 pixels. Main content remains visible with JavaScript execution disabled.
• Axe WCAG 2 A/AA and 2.1 AA checks found zero violations on the homepage in both light and dark mode; desktop and mobile screenshots reviewed. This is automated coverage, not a claim of complete accessibility conformance.
• No production dependencies or gateway behavior changed. No deployment performed.

Companion: Kimetsu implementation and public audit. The benchmark harness is maintained in a separate private repository.