case study

The site you are reading is the case study. It carries a scroll-scrubbed video hero, WebGL tools and an animated navigation — the exact features performance advice tells you to avoid — and still scores 99 on desktop Lighthouse. The pages without the video hero score 100. The one point is what the video costs, and it was worth paying. Verify every number from your own browser.

  • Lighthouse, desktop 99, with the video hero running
  • Lighthouse agentic browsing 3/3 checks passed
  • Design tokens ~80 custom properties, no framework

  1. The problem

    Marketing sites die of accumulation: a framework here, a tracker there, a carousel library — until the median business homepage takes seconds to become usable and gets quietly punished for it in search. The usual "fix" is a beautiful site that is slow, or a fast site that looks like a README.

  2. The constraint

    One engineer, one deadline, and a brief that refuses to trade away either half: the site had to look distinctive — a scroll-driven video hero, WebGL in the Lab, animated navigation — and still hit a Lighthouse floor most agencies only promise on an empty template. No performance team, no separate accessibility pass, no budget for a licensed video library: every constraint a real client brings, applied to the studio's own front door first.

  3. The decisions

    Astro renders everything to static HTML; JavaScript ships only for islands that are genuinely interactive, and loads only once they scroll into view. CSS is vanilla and built on design tokens, not a framework — around eighty custom properties drive color, spacing, type and motion sitewide. The heavy things are engineered around the metrics, not despite them: the hero video is scrubbed by scroll rather than autoplayed, composited from footage generated in-house rather than licensed stock; 3D loads lazily on its own route; every animation sits behind prefers-reduced-motion, which is not decoration — a visitor who asks for reduced motion never fetches the hero video's bytes at all.

  4. What each decision cost and bought

    Vanilla CSS instead of a utility framework cost more time upfront — eighty tokens had to be designed and named before the first component existed — and bought a site that gets re-skinned by touching a handful of variables instead of hunting through markup. A hand-built scroll-driven video engine instead of a plugin cost more implementation time and bought a hero with zero third-party runtime weight and full control over what reduced-motion visitors never download. Hero loops generated in-house cost production time on a local GPU instead of a licensing fee, and bought motion tuned to this exact page rather than stretched to fit it.

  5. Measured reality

    Desktop Lighthouse performance: 99, with the video hero running — not a stripped test build and not an empty template; the pages without the video hero measure 100 on the same run. Lighthouse's agentic-browsing audit — Google's own experimental check for how well a site reads to an AI agent — passes all three of its checks here: a well-formed accessibility tree, zero layout shift, and a valid llms.txt; check it yourself in PageSpeed Insights, or in Chrome DevTools if your Chrome is new enough — it takes a minute. The whole site ships in two full locales with correct hreflang. Mobile pays more for that hero: a scroll-scrubbed video is heavier on a phone's radio and CPU than a static one, and this design accepts a lower mobile number rather than trim the desktop experience — a trade named here, not hidden. Every number on this page was measured with Lighthouse; run it yourself and hold me to it.

  6. What is deliberately absent

    No CMS. Copy lives in code, so a wording change is a small pull request, not a dashboard edit — the trade a static-first, framework-free site makes, and one worth knowing before you buy it. No analytics suite loads ahead of the first paragraph; if a client needs numbers, they come from server logs or a script that does not cost Largest Contentful Paint. And no chasing the last few mobile Lighthouse points against a video hero — that fight was declined on purpose, not lost by accident.

What this proves for your project

That 'fast' and 'distinctive' is an engineering choice, not a budget tier. The websites package writes this same performance target into the contract because this site demonstrates it is deliverable, and the AI-search readiness clients buy separately as an audit — schema, llms.txt, crawler access — is built into every build from the first commit. How the contract, milestones and warranty actually work is on the Trust & Process page, and where this rung sits is on the full price ladder .

Ready when you are

Write two sentences about what you're trying to do. You'll get a straight answer — including "you don't need me for this" when that's the truth.

Start a conversation

Contract & NDA before any work beyond the audit · milestone payments, typically 30/40/30 · 30-day warranty · registered business, Armenia · GMT+4, EU hours · Trust & Process