When we started, the site had no SEO and almost no search traffic.
Source: the store’s own Shopify Analytics
The reusable parts every EcomOptix page is assembled from, rendered with the same stylesheet the real pages load. Open this before building a page, copy the pattern you need, and change the words — not the design. For logo, voice and identity rules, see the brand guidelines doc.
Hand-rolling a page from scratch is how the design drifts and how honest numbers turn into claims nobody can defend. Three rules cover almost every build.
home.src.html, a file in services/<slug>.json, or shared.css, then run python3 build.py. The builder refuses to write a page with an unfilled token, so half-finished pages can’t ship.shared.css as a token, never inline on a section.shared.css directly. If a component here looks wrong, the stylesheet is wrong — fix it there and every page follows.Every swatch below is drawn with var(--token) straight from shared.css. If you change a token, this page changes with it — it cannot drift.
Locked 22 Nov 2026. An audit of /ecommerce-seo/ found 16 greens, 7 light surfaces, 3 inks and 2 golds rendering on one page. The list below is now the whole palette. A colour that is not a token here does not ship.
Why #2a8019 and not #2f8f1c. #2f8f1c reaches only 4.14:1 on white — fine for a 70px headline (needs 3:1), a fail for a 13px label (needs 4.5:1). #2a8019 is the same hue (110°) and saturation, darkened until it clears AA everywhere: 5.00:1 on white, 4.72:1 on warm, 4.53:1 on mist. At display size the two are indistinguishable. --g-deep is retired and now aliases --g-core.
--cream (#f5f2ec) is retired and now aliases --warm. It was the fifth off-white on the page and nobody could tell it from the other four.
Ink carries the setup, green lands the payoff. It is the one piece of brand expression in the type system, so it has rules.
| Rule | Why |
|---|---|
| Green goes on the shorter half — at most ~40% of the headline | Colour is emphasis. When two-thirds of the words are green, the ink line reads as a leftover. |
| Green lands on the payoff, never the setup | The eye finishes on the coloured phrase. Put the thing you want remembered there. |
| Exactly one colour break per headline | Two breaks turn a headline into a list. |
Use <span class="g"> — never an inline colour |
It binds to --g-core, so it follows the token. |
| Label above and highlight below are the same green | Two near-identical greens 20px apart read as a bug, not a system. |
What the hero does today./ecommerce-seo/ puts 34 of 49 characters in green across two of three lines, and the kicker above it is a different green (#2f6f1b against #2f8f1c). Both are fixed by the rules above.
Two radii and one shadow. The audit found 14 radius values and 7 shadow recipes on a single page — including two different greens for the same hard shadow at two different offsets.
| Token | Value | Use it for |
|---|---|---|
--r |
18px | Cards, panels, media, dark CTA blocks |
--r-s |
14px | Number chips, tags, inputs, images inside a card |
--r-pill |
999px | Buttons and badges — the only other radius |
--sh |
7px 7px 0 var(–g) | The signature hard shadow. One offset, one green. |
--sh-soft |
0 8px 24px rgba(24,40,18,.06) | The only soft elevation on a light surface |
A circle (50%) is allowed for avatars, play buttons and icon dots. Nothing else may invent a radius. On hover the primary button lifts 2px and the shadow grows to calc(var(--off) + 2px) — that is the token, not a new value.
Two faces only. Barlow Condensed 900 all-caps for display, 800 for labels and buttons; Barlow 400 for everything you read. Body never drops below 17px and never uses weight 300.
Loaded weights: Barlow 400 / 500 / 600 / 700, Barlow Condensed 600 / 700 / 800 / 900. Do not write a weight outside that list — the browser will synthesise it and the headline goes soft.
| Role | Class | 390px | 1440px | Weight | Tracking | Case |
|---|---|---|---|---|---|---|
| Hero kicker | .hero h1 .kick |
13–14px | 15.8px | 600 | +1.3 to 2px | Upper |
| Hero headline | .hv-a / .hv-b |
42px | 70.6px | 900 | −1.2px | Upper |
| Section heading | .h2 |
32px | 48px | 900 | −0.5px | Upper |
| Card / sub heading | .h3 |
19px | 21.6px | 900 | +0.8px | Upper |
| Section label | .eyebrow / .lbl |
13px | 13.5px | 800 ← open | +2.4px | Upper |
| Lede | .lede |
17px | 18px | 400 | +0.1px | Sentence |
| Body | — | 17px | 17px | 400 | +0.1px | Sentence |
| Button | .btn-p / .btn-g |
16px | 19px | 900 | +0.8px | Upper |
These are measured, not intended. Every row above was read off the rendered /ecommerce-seo/ page at 390px and 1440px on 22 Nov 2026. Earlier versions of this table listed the sizes that were written — those rules lose the cascade and never reach the screen.
Only the headline is allowed maximum weight. The hero kicker ran at 900 — the same weight as the 70px headline below it — so it read as a small second headline competing with the real one rather than as a label. It is now 600 on the homepage and in shared.css (22 Nov 2026). On phones it also tightens to 13px / +1.3px tracking so the line does not wrap; that narrowing is deliberate, not drift.
Measure at an explicit width. Every size here is labelled 390px or 1440px because a computed value carries no hint of which media query produced it. Reading styles at whatever width a panel happens to be is how the kicker was first mis-recorded as 12px when it is 15.8px on desktop.
Why the scale drifts. The live service page carries 26 <style> blocks in the body on top of 15 in the head, because each past fix appended another block instead of editing the first. The hero headline now has ten competing font-size rules, six of them with no media query; the section heading has seven. The last one written wins, so the scale changes whenever anyone adds a block. Fix type by editing shared.css, never by appending a new style block.
| Token | Value | Used for |
|---|---|---|
--wrap |
1240px | Max content width on every section |
--gut |
clamp(16px, 5vw, 48px) | Side padding. Never let content touch the edge. |
--sec |
clamp(72px, 9vw, 128px) | Vertical padding on .sec |
--r |
18px | Card radius. Buttons are full pills (999px). |
--off |
7px | Hard offset shadow on bordered cards |
Roughly 40% of a page is dark. Light sections carry copy and proof; dark sections carry impact and the asks. Two darks never touch — put slate between them, or a light section. The pale-green band appears once, for the single biggest idea.
.sec alone is white. Add .paper (cream + dot grid), .warm, .mist, .slate, .dark, or .band.Each block below is rendered with the live stylesheet. Copy the markup, replace the words, keep the classes.
The wrapper every section starts from. .sec + surface + .wrap.
One or two sentences of context. Keep the line length short enough to read without effort.
<section class="sec paper"> <div class="wrap"> <span class="lbl">Section label</span> <h2 class="h2">A heading that makes the point, with <span class="g">one phrase highlighted.</span></h2> <p class="lede">One or two sentences of context.</p> </div> </section>
Pills, always. .btn-p is the one real action on the screen; .btn-g is the quiet alternative.
<a class="btn btn-p" href="#offer">Get my free growth analysis</a> <a class="btn btn-g" href="#results">See client results</a>
| Height | Size / weight | Radius | Fill | Shadow | |
|---|---|---|---|---|---|
.btn-p |
60px | 19px / 900 upper | --r-pill |
--ink, white text |
--sh |
.btn-g |
60px | 19px / 900 upper | --r-pill |
transparent, 2px --ink |
none |
One button, one size. The audit found 8 CTAs on /ecommerce-seo/ wearing 5 different looks — four heights (54 / 56 / 57 / 58px), three font sizes (14 / 17 / 19px), two shadow offsets (5px and 7px) and two radii. Two of them carried the class .btn-p and looked nothing alike: one was a black pill with a green hard shadow, the other a green 10px-radius block with no shadow. Never restyle a button inline or in a page-level block — if it needs to look different, it is a different component and belongs in shared.css.
A client number with its dashboard, dates and source. Never a number on its own.
When we started, the site had no SEO and almost no search traffic.
Source: the store’s own Shopify Analytics
<article class="card"> <span class="lbl">Client · market · platform</span> <div class="pw-big">$362,763</div> <div class="pw-lab">Total store sales, 19 May – 18 Jun 2026</div> <p>What the situation was and what we changed.</p> <p class="small pw-src">Source: the store's own Shopify Analytics</p> </article>
Short, concrete, in the buyer’s words. Lead each line with the bold claim.
<ul class="pains"> <li><span><b>The bold claim.</b> The sentence that lands it.</span></li> </ul>
One big idea, once per page, on pale green. No button, no list.
AI does the analysis and the volume. People own the strategy and the judgement calls.It’s how a focused team moves faster than a traditional agency, without handing your store to a bot.
<section class="sec band" aria-label="How we use AI"> <div class="wrap"><p>The one big idea.<small>The line that explains it.</small></p></div> </section>
Real objections only — cost, timeline, who does the work, where the team is. First one open.
The teardown comes first, so you can weigh the spend against the opportunity we actually found in your store, not against a guess.
Technical and conversion fixes can move numbers early because they work on traffic you already have. Commercial change usually shows between months three and six.
<div class="faq"> <details open><summary>The question in their words?</summary> <p>A straight answer.</p></details> <details><summary>Next question?</summary><p>…</p></details> </div>
Dark. One action, with the reassurance lines directly beneath it.
We look at your real store and your real competitors, then tell you what we’d fix first.
Free, no card · Reviewed by a person · Yours to keep
<section class="sec dark paper" id="cta"> <div class="wrap center"> <span class="lbl">Free growth analysis</span> <h2 class="h2">The promise. <span class="g">The payoff.</span></h2> <p class="lede">What actually happens next.</p> <a class="btn btn-p" href="#offer">Get my free growth analysis</a> <p class="small">Free, no card · Reviewed by a person · Yours to keep</p> </div> </section>
These orders are settled. Build to them unless there’s a reason to change, and change the recipe here first so the next page inherits it.
Proof sits directly under the hero — most visitors never scroll far.
Built from services/<slug>.json. Optional blocks drop out when the data isn’t there, so a page without a real case study can’t invent one.
One action only. No menu, no footer links, noindex. Booking sits high, not at the bottom.
.lbl, .eyebrow) are still 800 and should follow.shared.css first, so every page gets it..sec for its padding. The audit found the native sections at a clean 72/72 and the imported proof blocks at 65/70, 58/64 and 0/0 — the vertical beat stumbles exactly where the proof sits.#included numbered its steps with 56px black rounded-squares reading “01, 02, 03”; #offer used black circles reading “1, 2, 3”. Same page, same job, two designs — pick one and reuse it./ecommerce-seo/ carries 26 of them in the body on top of 15 in the head, which is why its hero headline has ten competing font-size rules. Edit shared.css and rebuild.Files:EMS/ecomoptix/v7/ — shared.css (the system), home.src.html, service-template.html, services/*.json, build.py. Identity, logo and voice live in skills/proposal-os/brand/brand-guidelines.html.
This page is noindex. Internal reference, not a public page.