lernerworks.com
The site you're reading, built the way I'd build yours. Every number on this page was measured, and the tooling that measures it is kept alongside the site.
In short
- The problem. My old site said I was careful and measurable without showing it: the work was three clicks deep and a placeholder was live on the homepage.
- What I built. A site that leads with terms and real work, loads in eight requests with nothing tracking visitors, and checks itself before every change goes live.
- What it means for you. Yours gets the same rules: it says what you do in the first line, loads fast on a weak signal, and every claim can be checked.
- Client
- Lerner Works, my own practice
- Sector
- Professional services, local
- Scope
- Positioning, structure, copy, design, build, link previews, search groundwork, performance, security headers, the tooling that checks all of it
- Year
- 2026, rebuilt September 2026
- Status
- Live. Re-measured after every change.
At a glance
- requests to any third party, on any page: no analytics, no font CDN, no widgets
- 0
- lines of JavaScript run in the visitor's browser, including the mobile menu
- 0
- to load the home page in full, images included
- 7 requests
- for the whole home page, of which 147 KB is images: the three case-study thumbnails
- 293 KB
The situation
A freelancer's website has one job: make a stranger comfortable enough to book a call. The first version of this site tried to do that with a headline, a "what a quiet phone costs you" calculator and a pricing table, and pushed the actual work to a secondary page.
Looked at the way I'd look at a client's site, it had four problems. The proof was hidden: the case studies were the strongest thing on the site and they were three clicks away. A placeholder card reading "PROJECT NAME" was live on the homepage. The calculator asked visitors to guess their own numbers and reported a made-up figure back to them, which is the opposite of credible. And on a phone, the navigation simply hid most of its links rather than offering a menu.
The problem in one sentence: the site asserted that I was careful and measurable, but didn't show it.
Constraints
No invented proof. No testimonials I chose myself, no client counts, no logo wall. Every project carries a label saying what kind of work it was, and unpaid work is never dressed as a commission. If a number couldn't be measured, it couldn't go on the page.
No third-party requests. No analytics script, no font service, no embedded booking widget, no chat bubble. A visitor's browser talks to this site and nothing else. That is a privacy position, but it is also most of what makes the pages fast.
No build step. Plain HTML and one stylesheet, so any page can be edited in a text editor and the site can be hosted anywhere for nothing. The dev tooling that exists (link-preview rendering, image re-encoding, measurement) runs on demand and ships nothing to the visitor.
What I built
A structure that leads with terms and proof. The homepage opens with the headline beside the five terms of working together: fixed price in writing, ownership of everything, one person, a refundable first design, and a reply within a business day. The three case studies come next, before the services. Work, Services, About and Contact each got their own page, so the home page could stop trying to be all of them.
<details> element, so it works with no script at all.One case-study template, the same six questions every time. Situation, constraints, what I built, decisions and tradeoffs, result, where it stands. Each study opens with a facts panel (client, sector, scope, year, status) and an "at a glance" strip that is limited to things that can be counted. A badge on every study says what kind of work it is: client project, own project, or this site.
How it's built
Type that is served from here. Source Serif 4 for headings and Public Sans for everything else, both variable fonts, both self-hosted. Only the Latin subsets are shipped, and the browser fetches the extended-Latin file only if a page actually contains one of those characters. English copy costs 76 KB of font, once; every page after the first reads it from cache.
Images that reserve their own space. Every image carries its real pixel dimensions, written by a small script that reads the files rather than trusting what was typed. That is what stops the page jumping as pictures arrive. Screenshots are re-encoded with the bundled browser rather than an image suite, and the cards use 800-pixel thumbnails instead of the full lead image.
The groundwork search engines look for. One h1 per page and no skipped heading levels. Canonical URLs, a sitemap that matches them character for character, and structured data describing the business and each case study. Link-preview cards for social platforms are generated from the site's own type and color, so a shared link looks like the page it points to.
A security policy that allows almost nothing. Every response carries a content-security policy that permits the site's own stylesheet, fonts and images and nothing else: no scripts from anywhere, no inline styles, no frames, no form posts. It is the "no third-party requests" claim enforced by the browser rather than promised by the page, and it is tested the same way, by serving the site with the policy on and checking every page for a single violation.
Checks that run before every deploy. The nav and footer on every page are rendered from one source by a script that either confirms every copy matches or rewrites it; the email and booking link are kept the same way. A second script checks every internal link and fragment, every heading level, every image's alt text and dimensions, every title and description for presence and uniqueness, and the sitemap against the pages. A third loads every page in a browser under the real security headers and runs an accessibility audit at three widths, with the mobile menu open and closed. All three live in the repository next to the pages they check.
Decisions and tradeoffs
No analytics on my own site. I set up call and form tracking for clients because their inquiries are the score, and I recommend it. Here, the score is whether the call gets booked, and the booking calendar already tells me that. So there is no analytics script, and the cost is that I don't know how many people read this page. I've decided that is a fair trade for a site that makes no requests I'd have to explain in a privacy policy.
The calculator went. It was the most interactive thing on the site, and it produced a number nobody could check. Removing it also removed the only JavaScript on the site. The pricing table stayed at the time and moved up; it came off later, for the reason in the next paragraph.
Prices quoted, not posted. Until October 2026 the site posted three fixed-price packages in full, on the argument that a visitor who had read the prices and still booked a call was a better call. They have since come off. Every project is still one fixed price in a written scope, agreed before work starts; the figure now comes after the first call rather than before it, and the cost is that a visitor can no longer check it before getting in touch.
No content management system. I am the only person who edits this site, so a CMS would be machinery to maintain for no one's benefit. For a client who needs to change their own text and photos, I make the opposite call, and the services page says so.
A case study about the site itself. I'd rather have a real study than a "coming soon" card. This one has the advantage that every decision can be shown, and every claim can be checked by reading the page you're on.
Result
Measured in headless Chromium at a 1,600-pixel viewport with an empty cache, using the script kept in the repository, after the rebuild. "Requests" is everything the page fetches; "external" is any request to a domain other than this one.
Every figure on this page is printed in one run of tools/measure.js --table and pasted in, so no two numbers can disagree.
| Page | Requests | External | JavaScript | Total |
|---|---|---|---|---|
| Home | 7 | 0 | none | 293 KB |
| Boulder County web design | 4 | 0 | none | 142 KB |
| Work | 9 | 0 | none | 431 KB |
| Inside the Towns | 27 | 0 | none | 629 KB |
| AragoCor Minerals | 7 | 0 | none | 595 KB |
| TownofNiwot.com | 8 | 0 | none | 929 KB |
| This page | 7 | 0 | none | 585 KB |
| Services | 4 | 0 | none | 144 KB |
| About | 4 | 0 | none | 136 KB |
| Contact | 4 | 0 | none | 136 KB |
Before the rebuild, the home page was 803 KB, almost all of it screenshots saved as PNG. Re-encoding them as JPEG and using an 800 px thumbnail on the cards rather than the full-size capture took it to 308 KB, with no change to what a reader sees. The same two moves are why the heaviest pages on the site are case studies carrying screenshots as evidence, rather than the home page. It is 293 KB now: the strip of counted figures under the hero and the stylesheet that draws it added a little, the lernerworks.com card's thumbnail is lighter than the one it replaced, and on October 11 the re-taken Inside the Towns thumbnail and the twenty-four town palettes the network diagram needs added 5 KB between them. The new logo added 5 KB more, to every page: it is drawn inline in the nav and inlined again as the favicon, so it is bytes rather than requests. On October 14, taking the pricing section off the home page, and its styles out of the stylesheet, took 4 KB off.
What is not measured here: PageSpeed scores, because they need a public URL and a moment in time, and real-visitor timings, because there is no analytics. The one Lighthouse figure elsewhere on this site, on the TownofNiwot.com study, carries its date and says where it was run.
What is checked rather than measured: every page passes axe-core at WCAG 2.2 AA at 1,280, 860 and 390 pixels, and loads under the content-security policy above without a single violation. Both are one command, tools/audit.js, run before every deploy.
Where it stands
Live, re-measured after every change, and checked before every deploy by the three scripts above, so the table is regenerated rather than trusted and a broken link can't ship. Each re-measure has caught something worth fixing, from out-of-date screenshots to images the script hadn't waited for.
What this means for your project
This is the site I would build for you, under the same rules: it says what it does in the first line, it loads fast on a weak signal, and every claim on it can be checked.
- Fast because it is plain. Hand-written pages, self-hosted fonts, no framework and no scripts, so a whole page is a handful of requests and loads on a cracked phone at the back of a job site.
- Yours, in full. Files a text editor can open, hosted anywhere for next to nothing, with no platform between you and your own site.
- Measured, not admired. The tooling that produced the table above ships with the site, and it would ship with yours, so what "fast" means is a number rather than an adjective.