Multi-model comparison page

A page that compares the same brief across three models: a header naming the brief, answer tabs (one per model) where each tab shows the model name, its recorded fidelity score, and its answer, plus a verdict callout that states the pick is editorial — fidelity scores are judgements from recorded runs, not benchmarks. Keep a source row under each answer. Keyboard-accessible tabs with aria-selected; focus is visible; colours are CSS variable tokens; reduced-motion keeps the tab underline static.