# Narration (first person): building GGQ UI demos v3 Generated (UTC): 2026-01-22 I started by refusing to touch CT250. The whole point of this v3 pass was: **new pages, new folder, no overlap** — so nobody’s production work gets clobbered. Then I pulled the *real* brief, not vibes: - the competitive landscape scan (what the best travel sites actually do) - and the 5 layout propositions (GGQ-specific constraints: ads must stay, search must lead, no bookings). From there, I made a decision that kept me sane: I wasn’t going to “redesign everything”. I was going to prove the *core interaction loop* first: 1) **Search-first** (type, get suggestions, hit enter) 2) **Destinations overlay** that truly supports **3 levels** 3) **Reading comfort** (article typography) 4) **Map credibility** (split view, not a tiny widget) 5) **Ads as first-class layout** (slots, labels, no overflow) I copied the v2 demo assets and demo data because it’s already grounded in GGQ reality (destinations tree, sample /system images, ad creatives). That saved time and reduced risk. The first delight was how quickly the new hierarchy snapped into place once I treated the header as a real tool: - sticky header - big search - destinations as a button (not a buried menu) The first frustration was the usual one: the moment you go “modern”, you can accidentally ship a11y regressions. So I baked the guardrails into the *base CSS* right away: - skip link - focus-visible ring - reduced-motion fallback Then I built the pages one by one: - **Home**: hero + “modes” tiles + a clean ad slot + regions + recent articles - **Destination**: planner-like layout with a leaderboard ad + right-rail ad (but never spilling) - **Article**: print-like measure + drop cap + captioned image - **Map**: split view mock (filters + results + big map area) - **Itinéraires**: a surface that points at the existing DB inventory report (so it’s not fantasy) When the pages existed, I ran the “don’t embarrass yourself” pass: - check that every search input has a label/aria-label + name - check buttons have explicit type - check all images have width/height That passed cleanly. Finally I did the one thing that actually matters: **screenshots**. If you can’t see it, you can’t judge it. Desktop + mobile renders let us compare v2 vs v3 without debates. At that point, v3 is ready to publish as a separate pack, and ready for a more serious “creative pass” with a design skill — because now we have a stable skeleton to push on.