Nobody scrolls sideways by accident

Every studio chasing an award eventually arrives at the same idea: what if the site did not scroll the normal way? Horizontal scroll. Scroll-snapped chapters. A canvas you drag. A radial menu. A timeline you scrub. The reasoning is sound on the surface — juries explicitly reward innovation, and nothing announces innovation faster than breaking the most familiar interaction on the internet.

It also has an extremely high failure rate. We have watched beautiful, expensive, structurally unusual sites score 5 on usability and lose to something far more conservative. The pattern is consistent enough to state as a rule.

The single question

Does the unconventional structure come from the content, or from the desire to be unconventional?

That is the whole test, and it predicts the outcome with uncomfortable accuracy.

An archive of a photographer's forty-year career, presented as a horizontal timeline you travel along, is structurally motivated. The content *is* a sequence through time; the interaction is a physical metaphor for the subject. A judge encounters it, understands it immediately, and reads it as insight.

A B2B software company's homepage — hero, three features, pricing, testimonials — rotated ninety degrees so you scroll sideways is not structurally motivated. It is the same information architecture with an obstacle attached. The judge understands it immediately too, and reads it as decoration, then knocks points off usability for the trackpad friction.

Same technique. Opposite score. The difference was never the execution.

What breaks, specifically

If you do commit to an unconventional model, these are the things that quietly destroy it. Every one of them is fixable, and almost none of them get fixed.

The scroll contract. When you hijack the scroll wheel, you have taken away a control the user has used ten thousand times this year. If the mapping is not one-to-one — if a wheel tick produces a variable, eased, decelerating movement — the user's sense of direct manipulation is gone. Hijacked scroll must be *immediately* responsive and must never overshoot. The moment a user pushes and the page keeps moving after they have stopped, they have lost the connection.

Trackpads and mice are different instruments. A mouse wheel produces large discrete jumps. A trackpad produces a continuous stream of small deltas with momentum applied by the operating system. An interaction tuned on a MacBook trackpad frequently feels violent on a mouse, and vice versa. Test both, on both platforms, before you show anyone.

Keyboard access. This is the most common outright failure. Custom scroll containers routinely break arrow keys, Page Up and Down, Home and End, and tab-to-focus scrolling. If a keyboard user cannot traverse your site, you have failed an explicit accessibility criterion — and on the platforms that publish a rubric, accessibility is a scored category, not a footnote.

Deep linking. Can a judge send a colleague a URL that lands on a specific piece of work? If every state on your site lives at the same address, you have built an app that forgot it was a website. Sync state to the URL.

Browser back. Users press it constantly. If your custom navigation does not push history entries, back exits the site entirely, and the visitor does not come back.

The mobile translation. Horizontal scroll on a touchscreen conflicts with the browser's own gestures and with the user's mental model of a page. The honest answer is usually that the mobile version becomes a well-designed vertical sequence. That is not a failure — but the vertical version must be *designed*, not generated by letting the desktop layout wrap.

An unusual interaction has to be learnable in one gesture. If it needs a tooltip explaining how to move, it has already cost more than it earns.

The cheaper alternatives that score just as well

The innovation column does not actually require you to break scrolling. It requires the judge to encounter something they have not seen arranged this way before. That can come from structure, sequencing or material rather than from input mechanics.

Chapter architecture. Vertical scroll, entirely conventional, but the page is organised as numbered chapters with a persistent index that updates as you move. Familiar to operate, unusual to read, and it makes long content feel navigable.

Sticky comparison. One side of the composition holds while the other moves, so two things can be read against each other. Enormously effective for before/after, problem/solution, or claim/evidence — and it uses nothing but position: sticky.

A page that behaves like an object. A site organised as an invoice, a contract, a filing card, a menu, a specimen sheet. The metaphor does the innovative work; the scrolling stays boring, which means it stays usable.

State as the surprise. Nothing unusual about the navigation at all, but the page transforms — inverts, recomposes, changes register — at a moment of narrative significance. One well-placed transformation is more memorable than a hundred scroll-linked parallax layers.

Sound, used once. Almost nobody does audio well, which is exactly why a single, optional, well-designed sonic moment is more distinctive than another cursor effect. Default to off, always.

The version of this we believe in

Unconventional structure should be the *consequence* of taking the content seriously, never the starting brief. When we find ourselves saying "it would be cool if it scrolled sideways" before we have read the client's material, we stop, because we have started designing the award entry instead of the website.

The sites that win both columns — innovation and usability — nearly always share one quality. Their strangeness is legible. You arrive, you are surprised, and within two seconds you understand both how to operate the thing and why it was built this way. That legibility is the craft. The strangeness is easy.