00b · client demonstrated, we inferred · the system every page here obeys

We measured the brand before we designed anything.

We were not given brand guidelines for this brief, so we derived a working set from the live site: three pages sampled element by element, then the rules behind the values written down. helfy® may well hold guidelines internally; these are ours, and everything in this walkthrough is built from them.

How we derived it

Measured, not eyeballed.

A style guide written by looking at screenshots is a guess. These values came out of the browser itself, which is why they include things nobody would invent, like a 10.5px root unit.

01Ran it through /taste. A Claude Code skill built for this: it drives a real browser to the target, screenshots it, and walks the rendered DOM. behelfy.com was loaded at 1440×900 in a clean profile, so no cookie or login state could change what rendered.
02Sampled every rendered element. Computed styles from the whole page rather than the stylesheet: colours ranked by the share of visible surface they actually cover, type by frequency of use, spacing by what repeats. It also picks up the things nobody writes down — letter-spacing per role, line-height ratios, the exact geometry and tint of a drop shadow, whether :focus-visible and prefers-reduced-motion are authored.
03Treated the screenshot as ground truth. Where CSS and the eye disagreed, the eye won. A colour present in the DOM but covering under 5% of the page is decorative, not a brand colour. Imperceptible shadows are artefacts, not decisions.
04Cross-checked three pages. Home, a product listing and the blog. A value appearing on two or more pages is a system rule; one page only makes it local. That distinction is what stops a one-off being enshrined as a standard.
05Separated the platform from the brand. It read the site's own CSS custom properties, which is how we know these are Shopify pages on a Dawn-derived theme: --font-body-scale of 1.05, --page-width of 120rem, --spacing-sections-desktop of 36px. That is why the numbers carry odd fractions, and why some of them are theme settings rather than design decisions.
06Wrote down the trade-offs, not just the values. Tokens tell you spacing is 10.5px. They do not tell you why the accent never appears on a heading. The rules below are the part that lets someone design a new page that still belongs.
Colour · measured by share of visible surface

Two grounds, one accent.

Orange covers between 0.1% and 1.6% of any page. That is not restraint for its own sake: it means the next action is findable by colour alone, with no competing orange to disambiguate.

Teal#1A4747Full-bleed section surface
56.5–71.4%
White#FFFFFFPage ground
24.7–37.3%
Orange#FF751ACTA fills only
0.1–1.6%
Pink#FFD1D1Soft band, footer
2.2–5.8%
Black#000000Headings
the only fixed text value

There is no grey in the system. Every secondary tone is black or white at reduced alpha — rgba(0,0,0,0.75) for body copy — so it borrows the hue of whatever sits behind it. One text token works on white, on teal and on the pink footer without an override.

Type · two weights, no middle

Montserrat everywhere. Mono on the buttons, and on the machinery.

Only 400 and 700 exist. No 500, no 600, across three pages. Hierarchy is built from size and colour alone, so a reader is only ever reading or being told the topic.

Know your health.

Hero · 52px · 700 · lh 1.30 · +0.6px

Take your personalised test.

Heading · 40px · 700 · lh 1.30 · +0.6px

The truth about sleeping tablets

Card title · 24px · 700 · lh 1.30

Quick links

Sub-head · 18px · 700 · lh 1.30

Quit taking random supplements and guessing what your body needs.

Lead · 16.8px · 400 · lh 1.76 · +0.63px

Get started

Button · 15.75px · Roboto Mono · +1.05px

Receive a detailed report breaking down your results.

Body · 14.7px · 400 · lh 1.76

BalanceKit

Product title · 13px · 700 · lh 1.30

© 2026, helfy · Refund policy · Privacy policy

Meta · 11.55px · 400

04 · ADS → QUIZ

Label · 11.55px · Roboto Mono · +1.05px · this deck's extension

The sizes are a rem ladder on a 10.5px root, not a modular scale — measured steps are 1.07×, 1.30×, 1.33× and 1.67×. To extend the system, replicate the ladder, not a 1.25× progression. One honest deviation: on helfy's own site mono is reserved for button labels, and every row above except Button is Montserrat. This deck extends mono to eyebrows, labels, trace lines and email meta, because a walkthrough of a routing machine wants its instrumentation to read as instrumentation. It buys a documentary register and spends part of what mono was doing for the buttons.

Space & geometry

Everything is a multiple of 10.5.

A single 1.05 scale factor on a 10px root produces the type ladder, the spacing ladder and the container width. The odd fractions are a consequence of that one number, not hand-picked values.

Base unit10.5px
Section gap36px
Container1260px
Side padding52.5px
Media radius12px
Button radius40px (pill)
Button height47px
Grid gutter21px
No dividers, anywhere

There is not a single horizontal rule anywhere on the site. Sections are separated by a full-bleed background change at a 36px gap. The colour band does the work a divider line would.

Components · the pill and the containerless card

A button looks like machinery.

These are the two components the whole system turns on, rebuilt from the measured values and shown at the sizes the site actually uses.

Button · 47px tall, padding 0 31.5px, radius 40px
Get started
Add to cart

Mono label at 15.75px with 1.05px tracking, the widest tracking anywhere in the system. Primary is #FF751A with a black label, not white. The secondary variant only exists on teal.

Card · no fill, no border, no shadow
BalanceKit£470.00

The image is the only edge: 12px rounded on the top two corners, flush along the bottom so the crop meets the text. Content centred. A row of these reads as objects rather than boxes.

Elevation · one layer, tinted, mostly absent

Left: the only shadow in the system, 10px 12px 20px at alpha 0.1, tinted rgba(26,71,71,·) so it reads as light in a teal room rather than a generic drop shadow. Right: what most surfaces actually get. The blog page carries none at all.

The rules behind the values

Four trade-offs worth inheriting.

Each of these is a decision with a plausible alternative that another designer would reasonably have taken. That is what makes them useful rather than descriptive.

Personality is spent entirely on the button

The decision

A second typeface used only on button labels, over the conventional display-font-for-headlines pairing.

Why it holds

On a page selling blood tests, a headline in a novelty face reads as marketing; a label that looks like terminal output reads as a control. People trust a switch that looks like machinery.

Orange is a verb, never an adjective

The decision

Accent confined to fills on clickable things, over spreading it into headings, icons, links and hover states.

Why it holds

Someone deciding whether to spend £499 needs to find the next step without first working out which orange thing is the button.

The photograph is the card

The decision

No fill, no border, no shadow on cards — a 12px-rounded image with a flush bottom edge is the only boundary.

Why it holds

When the product is a physical kit that arrives in the post, the buyer wants to look at the object, not at a box drawn around the object.

One text colour for every ground

The decision

Secondary text built from black-and-white alpha, over a grey ramp with per-surface overrides.

Why it holds

Text that inherits its surface never looks pasted on. A fixed #666 reads cold on the pink footer and muddy on teal.

Caveats, stated plainly

What is brand, and what is theme default.

Some values are Shopify settings

behelfy.com runs a Dawn-derived theme. The 10.5px root, the 1260px container, the 36px section gap and the radius values are theme knobs. The fractional pixel values are a consequence of the 1.05 body scale, not hand-picked numbers.

The brand-level decisions are elsewhere

The teal/orange/pink split, the monospace CTA, the two-weight ladder and the containerless card are where the actual taste lives. Those are the parts worth defending in a review.

Retrofitted after the day

On the hackathon day the time went on strategy and the funnel framework, which is the part that has to be right. This design system was measured and written up afterwards. Design is window dressing that can change after the build; the framework underneath it cannot.

This is our reading, not helfy's document

None were supplied with the brief, so this is what the live site does, measured and written down. It should be checked against anything helfy® holds internally before it is treated as canonical.

Applied to this deck

The walkthrough you are reading is the proof.

The original version of this deck was warm cream with a serif display face — a look with no relationship to helfy. Every page here was rebuilt on the system above: white and teal grounds, Montserrat in two weights, mono pill CTAs, containerless cards, and colour bands instead of divider rules.

The kit was produced with /taste, a Claude Code skill that measures a live site and writes the system out as tokens plus rules. Those rules were loaded into this build, which is how seven pages stayed on-system: two font weights, one accent, and no divider line anywhere.

Never Guess®. The system follows from that.

Brand guidelines were discovered and rebuilt after the day, not during it. Design system measured from behelfy.com across three pages at 1440×900; values are observed, not supplied by helfy®. Verify against any internal brand documentation before treating as canonical.