What changed, and why

The first version opened with three abstract claims and a three-column proof strip. Both were assertions: "we publish what we exclude" is the brand telling you it is trustworthy. They have been replaced by a label reader — a real ingredients panel with the failures marked and the reason attached. Same argument, shown instead of stated, and it doubles as the clearest possible preview of the product.

Four destinations, no sub-levels

Home, The Standard, Guides, Our Dogs — plus a persistent Join action styled as a button rather than a nav item, because it is a task and not a place. No second level, no dropdowns, no breadcrumbs; a site this size does not earn one.

The Standard is second, where About would normally sit

Nav order encodes priority and the standard is the entire proposition.

One primary action per view; no dead ends

Every view ends with exactly one call to action — always the waitlist. Secondary links are outlined rather than filled so the two never compete. Two equal buttons is the most common way a page converts nothing.

Progressive disclosure

The label reader shows six flags but only one reason at a time. The standard shows six named exclusions with the reasoning folded away. Full transparency presented all at once is less readable, and unreadable transparency persuades nobody.

The pet component

Each dog is drawn from their own coat colours and markings — mask, saddle, eye colour — rather than being random fur. It exists so the page works before photography does, and so the same component can render a customer's pet from the coat field FR‑R11 adds. It is a placeholder for real photographs, not a substitute for them.

The hero gesture

Two earlier attempts failed the same way. Both leaned on a solid base wash to fake density — and a fill is a background image. No quantity of strokes drawn on top of one will read as fur, because the eye sees the flat block first.

There is no fill now. Density comes from the only place it can: hundreds of thousands of individual overlapping strands, up to 420,000 depending on the size of your window. That is far too many to redraw sixty times a second, so the field is rendered once — progressively, over several frames, which is why the coat visibly thickens as the page settles — and then warped. The coat is blitted as a grid of tiles, each pushed by a spring displacement field.

So dragging moves every strand rather than a sparse top layer, and the gaps that open expose the pale undercoat beneath. That is what parting real fur looks like, and it is the brand thesis performed rather than described: the layer you only see when you part the fur.

The coat waves. A baked image cannot, so one strand in five is left out of the bake and redrawn every frame — a slow travelling wave, plus whatever your hand is doing. Roots stay pinned; only tips move, which is the one property that separates hair from cloth.

Under prefers-reduced-motion the field builds in one pass, the wave is switched off, and nothing animates or responds.

This breaks R2, deliberately

The brand system says motion is a response and FR‑R6 forbids ambient animation outright. A continuously waving coat is exactly that, and it is here because the reference for this hero was a stock clip of real fur moving in slow motion.

It is defensible in one place only — this hero, above the fold, where the job is to establish what the brand is made of. It is not defensible on a product page, in the app, or behind the standard. If it stays, R2 needs an explicit carve-out naming the hero, rather than being quietly ignored.

On the reference footage

The clip is licensed Shutterstock footage, and the downloaded file is the watermarked preview. None of it is embedded here and none of it should be: the fur on this page is generated, so it is ours, it is about 40 KB of code rather than megabytes of video, it recolours with the palette, and it responds to the cursor — which footage cannot.

If the real clip is wanted, it has to be licensed first. Same call we made on the tick illustration.

The fur-heavy treatment — and what it costs

This variant deliberately exceeds the dosage ceilings in §07 of the brand document, and by a lot. Grain runs at 13% in light and 20% in dark against a documented ceiling of 5% — roughly three to four times — plus a full-bleed live coat, shed bands between sections, and coat-filled card headers.

The texture is also drawn differently. Earlier passes grew long strands up from the bottom edge, which reads as a crop field rather than a coat. Fur is short, dense, overlapping, and lies along a direction that drifts across the surface, so strokes are now scattered across the whole area and angled by a smooth flow field over a base wash.

Two consequences worth being explicit about:

• The document and the site now disagree. If this treatment is kept, the dosage table has to be updated to match — a design system nobody follows is worse than no design system, because it makes every later argument unenforceable.
• The Standard stays at zero texture regardless. That rule did not bend: it is a credibility surface, and it is the one page here doing the persuading.

To revert: git revert the commit, or set --grain-op back to .05 and drop the .coat-hero and .band blocks.

Honest limitations

• Hash-routed prototype. The real site must be server-rendered and indexable (FR‑M14) — content discovery is the whole acquisition mechanism.
• The waitlist form is inert.
• The standard's exclusions are illustrative and unsigned.
• The label panels are realistic but invented; use real products only after legal review, and never name a competitor as rejected.