This page explains how this demonstration site is built and how it was tested. Cedar Harbor is a fictional city.
This website is a product demonstration. It shows a complete city government website, filled with the content of a city that was made up for the purpose. No real city, staff member, seal or photograph is used anywhere, and no person is named.
Who made it
The product is called TDG Civic. It is made by THATDEVELOPERGUY LLC of Cassville, Missouri.
- Email: admin@thatdeveloperguy.com
- Phone: 417 671 2606
- Website: thatdeveloperguy.com
What it is built with
The site runs on WordPress, using the editor that comes with WordPress and nothing added on top of it. It has two parts made for this product: a theme, which is the design, and a plugin, which is everything else. Keeping them apart means a city’s content survives a change of design.
- No page builder.
- No other plugins.
- No jQuery on any public page.
- Fonts are served from this site, not from another company.
- Nothing is loaded from another website, except a map, video or form from a source the city has approved.
- Every page is complete without scripts. Scripts only add convenience.
Guardrails for the people who edit it
A city website is edited by busy staff, not by web designers. The product is built so that a staff member can publish a correct page without help, and cannot break the design while doing it.
- A starting layout for every kind of content. A new service, event, document or news story opens with the right headings already in place.
- A short list of building blocks. Editors get what a city page needs. They do not get the tools that break layouts on phones or let unsafe code onto a page.
- The design is locked. Editors cannot change colors, fonts or spacing on a page, so every page looks like part of the same site and every color pairing stays readable.
- A publishing checklist. As an editor writes, the checklist lists anything that would shut some residents out: a picture with no description, a heading that skips a level, a link that says only “click here”, a table with no header row, a PDF that a screen reader cannot read. Each item says what to do and has a button that goes to the spot.
- Errors block publishing. A page with an error on the checklist cannot be published. The server enforces this, not only the editor, so the rule holds however the page arrives. A draft can always be saved, so no work is lost.
- Alerts stay short. An alert is one locked paragraph, with a level and a time to stop showing.
- Embeds from approved sources only. An administrator lists the websites whose maps, videos or forms may be shown. Nothing else can be embedded, and every embed must have a title.
How accessibility was approached
The product was built to the Web Content Accessibility Guidelines, version 2.2, at level AA. We do not claim that it conforms. Whether a city’s finished website conforms depends on the content that city adds, and is for the city to say after its own audit. What we can say is what was done and what was tested.
- Pages are built from plain, meaningful HTML: real headings, lists, tables, buttons and landmarks.
- Everything works with a keyboard, and the place you are is always visible.
- Every pairing of text and background is checked by calculation, in the standard colors and in each alternative color scheme.
- An alert that is on the page when it loads is a labeled region. An alert that arrives later is announced, and only an emergency interrupts.
- A form lists every problem at once, moves you to the list, links each problem to its field and keeps what you typed.
- Documents are web pages first. A file is an extra, and a PDF is checked for the structure a screen reader needs before it can be published.
- Motion is small, and is switched off for anyone whose device asks for less of it.
What was tested, and what was found
| Test | How it was done | Result |
|---|---|---|
| Rules that must not fail | More than 2000 automated tests of the rules behind alerts, events, forms, embeds, records, search and the publishing checklist | All pass |
| The site in real browsers | Automated tests in Chromium, Firefox and WebKit, and at phone and tablet sizes, carrying out the tasks a resident and an editor would | More than 690 checks pass, none fail |
| Automated accessibility checks | The axe engine on every kind of page at desktop and phone size | 0 failures in 50 page checks |
| Keyboard | Automated tests press Tab through the home page, open the menu, page the calendar, dismiss an alert and send a form | Every stop shows where you are, and nothing traps the keyboard |
| What a screen reader is given | The names, roles and states the browser hands to a screen reader were recorded and checked on six key pages | 64 checks hold |
| Zoom, wider spacing and larger text | Every kind of page was loaded at the width of a window zoomed to 400 percent, with wider spacing between lines, letters and words, and with text at twice its size | Nothing scrolls sideways and no text is cut off |
| Scripts switched off | The main tasks were repeated with scripts off | All work |
| Links | Every link on the site was followed | 112 addresses, none broken |
| Setting it up as a different city | A copy with no demonstration content was turned into the website of another made up city, twice, by following the setup guide in the dashboard | The first time found 16 faults in the guide and 8 in the product. After they were fixed, every step of the guide worked as written. |
| Holding still while it loads | Nine pages were loaded from a cold start with their pictures, and then their fonts, held back for most of a second | Late pictures move nothing. Late fonts move a line by a pixel or two. |
| Speed on a slow phone | Measured in a laboratory profile of a mid range phone on a slow connection | Main content appears in about 1.5 seconds at the slowest. A tap is answered in under a tenth of a second. |
The tests found real problems while the product was being built, and each was fixed. For example: focus was lost after a form was sent, the menu button did nothing on a phone with scripts off, a link in the trail above each news story led nowhere, and an alert could be published with its message typed into its headline.
The table gives the results for version 1.0.0. Version 1.0.1 changed one thing in the product: the site no longer announces which version of WordPress it runs. The automated tests were run again on version 1.0.1 and none fail. Setting the product up as a different city and the speed measurement were not repeated.
What was checked on the public server
The public copy of this demonstration runs on a real web server, at cedarharbor.thatwebhostingguy.com. On November 21, 2026 the checks that are safe to run on a site the public is using were run against that copy, running version 1.0.1.
| Check | How it was done | Result |
|---|---|---|
| Every kind of page | 25 pages were opened in a browser at desktop and phone size | 50 page loads. None had an error, a failed request or a request to another website. |
| The site in real browsers | The tests that only read pages were run in Chromium, Firefox and WebKit, and at phone and tablet sizes | More than 440 checks pass, none fail |
| Automated accessibility checks | The axe engine on the same 25 pages at desktop and phone size | 0 failures in 50 page checks |
| What a screen reader is given | The names, roles and states the browser hands to a screen reader were recorded and checked on six key pages | 64 checks hold |
| Scripts switched off | With scripts off, the menu, search, alerts, the calendar, the document library and the expandable sections were used. The request form was sent empty, then filled in. | All work. The empty form came back with its list of problems, and the second request was stored. |
| The request form | A request was sent the way a visitor would send it, then looked for in the dashboard | Stored, given a reference number, and listed for staff |
| Links | Every link on the site was followed | More than 100 addresses, none broken |
| The version number of WordPress | 36 addresses were read, among them every kind of page, the sign in page and the feeds | None announces the version |
| What must stay closed | The settings file, the installer, the list of users and other closed addresses were asked for, and each security header was counted | Each closed address refuses, and each header is sent once |
What has not been tested
It is as important to say what was not done.
- Nobody has yet listened to the site with a screen reader. The information a screen reader needs was checked, but how the site sounds was not. This is the most important test still to do, and a script for it comes with the product.
- The speed figures come from a laboratory test on one computer. They are not measurements of real visitors.
- On the public server, the tests that sign in as an editor, change settings or read the database were not run, because a site the public is using should not have that done to it. Those tests were run on a local test site.
- The product has run on one real server, with the nginx web server, PHP 8.4 and an SQLite database. It has not been tested with MySQL or MariaDB, behind a page cache, or with many visitors at once.
- Mail leaving a real server was not tested. The public server sends no mail.
- No security specialist has tried to break into the site. Not announcing the version number of WordPress is a small precaution. Installing updates promptly is what protects a site.
- Real phones and tablets were not used. The tests used browser profiles of them.
About this demonstration
- The City of Cedar Harbor does not exist. Phone numbers, email addresses and web links on this site are reserved for examples and reach nobody.
- The forms work. What you send is stored on this demonstration site and nobody will act on it. Use made up details, not your own.
- This site measures nothing about its visitors. It has no analytics.
- This site asks search engines not to list it, because it describes a city that is not real.
- The pictures were drawn for this demonstration. No stock images are used.
