Observations:

• Transfer size is consistently in the 330-390 KB band. Healthy for a full-article page; the theme / framework inflates about 280 KB of this.
• First-byte latency is under 200 ms (via server); full HTML arrives well under a second.
• Scripts per page are 48-50 — mostly inlined (WebLLM bubble, DYK widget, verified-badge JS, breadcrumbs, share-bar, surprise-me, gamification). No external JS calls apart from Google Fonts and AdSense.
• Image count is bounded; the explanations/grounds-for-rejection page has 45 images because of the 15 clause icons (SVG, tiny) + infographic diagrams.

• LCP (Largest Contentful Paint) — most likely the hero banner (~40 KB PNG). Lazy-loaded below the fold on most articles. Target <2.5s (good).
• CLS (Cumulative Layout Shift) — low risk. The floating chat bubble is appended after DOM-ready, well outside the viewport centre. The DYK widget is sidebar, fixed width.
• INP (Interaction to Next Paint) — the main risk area is the DYK widget button (simple DOM swap, negligible) and the floating-chat iframe opening (iframe load time dominates on first click).

1. Run Lighthouse on each of the top-20 pages via headless Chrome. Record scores for performance, accessibility, best-practices, SEO.
2. Sample with Chrome UX Report (CrUX) for real-user monitoring once traffic is sufficient (currently below the 10k-URLs-per-origin threshold).
3. Use PageSpeed Insights API to measure the same set monthly; record in a spreadsheet or dashboard.
4. Isolate scripting cost — profile the breadcrumb + surprise-me + DYK + chat launcher stack under CPU throttling. Combine into a single bundle if any one crosses 50 ms on mid-range devices.
5. Image optimisation — auto-generated hero banners are already WebP-ready; explicitly specify width/height on `<img>` tags to prevent CLS.
6. Font loading — Google Fonts is the only external font; consider `font-display: swap` to eliminate FOIT.

• SVG icons instead of raster for the grounds-for-rejection clause list — 45 icons totalling ~6 KB.
• Lazy iframe for the floating chat — does not load /tools/chat-app.html until the bubble is clicked.
• Deferred script loading — the WebLLM library is dynamically imported only when the user opts in to deep analysis.
• Preconnect hints — already in place for Google Fonts in pageheader.html.
• Cache headers — Apache mod_expires already set for images/CSS/JS to 1 year (configured in /home/bighelpers/public_html/rtiwiki/.htaccess).

• AdSense adds ~30-50 KB of deferred script. Each ad slot triggers a separate render call. Mitigate by keeping ad density low.
• Google Translate widget is on-demand (click-to-load) — zero cost until activated.
• DokuWiki legacy CSS includes several unused selectors; a CSS purge pass could trim ~20 KB.

• Measurement methodology: curl with `–write-out` template for transfer size and timing.
• Core Web Vitals thresholds: Google's official guidelines at web.dev/vitals.

Last reviewed: 21 April 2026. Baseline snapshot — full Lighthouse audit pending.