What we measured, and what we have not.

Where this site stands

We aim to meet WCAG 2.2 Level AA across this site. As of 24 October 2026, automated testing finds no violation of it on any page.

We are not claiming the site conforms. Conformance is a statement about the whole standard, and a good half of it can only be judged by a person — so claiming it on the strength of a tool would be inventing the half nobody tested. The site has not yet been tested by people who use assistive technology day to day. When it is, this page will say so and say what they found.

The figures below describe the automated audit of this build on 24 October 2026. The next build runs its own checks, and a failed audit stops it.

What was measured, and how

What was tested

Every one of the 155 pages in the build, on both sites, at 390 px and 1440 px — 310 runs in all.

Against what

WCAG 2.2 Level A and AA. Those are the rules the standard sets out; the tool checks the ones a machine can check.

With what

axe-core 4.14.0, an open-source testing engine, run from the build itself rather than once by hand.

When

24 October 2026. It runs again on every change, and the build fails if anything breaks.

Result

No violations, on any page, at either width.

The same build also checks color contrast on the palette before a page is ever drawn, refuses to publish a page that scrolls sideways at 320 pixels, and refuses one with a control too small to hit. Those run on every change, not once a year.

What it found, and what changed

The first run failed, and so did the run after it. Five real problems, all now fixed — they are listed because a statement that only reports the pass is not worth reading.

• Footer links were marked out by color alone

  The links along the bottom of every page sat inside a line of muted text and were told apart only by being lighter. That fails WCAG 1.4.1 — anyone who cannot distinguish the two shades cannot see there is a link there. They are underlined now.

• A phone number on the paid thank-you page was unreadable

  It used the link color tuned for white paper, on a near-black panel: a contrast ratio of 2.77 to 1 where 4.5 is the minimum. Links on dark surfaces now use the accent the rest of the dark design already uses, measured at 7.42 to 1, and they are underlined too.

• Five pages carried the same internal name twice

  Two drawings on one page were emitting identical ids for their gradients and filters. That is invalid, and it lets one drawing reference another’s definition. Every drawing now gets a name of its own, and no page in the build repeats an id.

• A primary button on dark panels was unreadable

  White text on the tangerine of the “Call” button — 2.33 to 1 — because the rule coloring links on dark panels outranked the one coloring buttons. It had been that way since the design system was first applied. Nothing caught it because the audit had been looking while those panels were still fading in, so the button counted as invisible and was skipped. The audit now forces the finished state, and it appeared immediately.

• One line of the site belonged to no region

  The bar across the top of every page was content sitting outside every landmark, so somebody moving through the page by landmark would never reach it. It is a labeled region now.

Earlier testing flagged the “skip to content” link because it sits before the page’s regions. That placement lets keyboard users reach the main content first.

What this kind of testing cannot see

Automated testing catches somewhere between a third and a half of what actually gets in someone’s way. Here is what it cannot tell us:

• Whether the description of a picture is accurate. A tool can see that a description exists. It cannot see that it is right.
• Whether the order things are read in makes sense once the layout is gone.
• Whether somebody using a screen reader can actually get through the estimate form and out the other side.
• Whether a heading describes what is under it, or a link says where it goes when read on its own.
• Whether something that moves can be stopped. The rule for that is not one a tool can check, and ours does not have it — we found a scrolling strip with no Pause button that way, by looking, not by running the tool.
• Whether any of it works for somebody with low vision, a tremor, a migraine, or a cognitive difference — which is the only test that finally matters.

So the honest position is that this site has been checked, not proven. If you hit something, you are finding what the tool could not.

What is built in on purpose

Skip to content

The first thing keyboard focus reaches on every page is a link straight past the navigation.

Everything is reachable by keyboard

Every control is an ordinary button, link or form field. Focus is drawn with a three-pixel outline that changes color on dark panels so it stays visible.

Targets are big enough to hit

A build check measures every control on every page at eighteen widths and on a landscape phone: nothing under 24 pixels, and nothing under 44 for a primary control.

It reflows instead of scrolling sideways

The same check fails the build if any page scrolls horizontally at 320 pixels, which is the width the standard sets for reflow.

Motion is optional

If your system asks for reduced motion, the reveal animations and smooth scrolling do not run.

Nothing moves that you cannot stop

There is no video and no audio. On our advertising pages a strip of brand marks scrolls, and there is a Pause button beside it. Everything else settles once as the page opens and then holds still. If your device asks for reduced motion, none of it moves at all.

Real text, not pictures of text

Every word on the site is text you can select, search, translate and enlarge. The page language is declared so a screen reader pronounces it correctly.

The form says what it wants

Every field has a visible label, errors are announced where they happen and say what to do, and no field relies on a placeholder to explain itself.

If something here is in your way

Tell us, and we will fix it. Email contactus@heatherwindowanddoor.com or call (469) 281-7750.

Say which page it was and what happened — “the form would not let me past step two with a keyboard” is more use to us than a standard reference, though a reference is welcome too. We read them and we fix what we can. We are not going to promise you a date, for the same reason there is no arrival window promised on the phone for a broken door: a deadline we miss is worse than no deadline.

You never have to use this website to get service. A real person answers the phone 24/7, and everything the site does — describing the problem, getting a written estimate, booking the visit — can be done by phone instead, on 7 AM–11 PM, seven days or outside them. If the website is the barrier, go around it: (469) 281-7750.

Describe it however makes sense to you

The form, if you would rather type than talk.

Every field has a label, every error says what to do, and the whole of it works with a keyboard. If any of that turns out to be untrue, the number above is the faster route.

• However you describe it is enough. No part names needed.
• The written estimate is free. Nothing owed if you pass.
• Nobody starts the repair until you approve the number.