Are there circumstances in which an overlay tool could be seen as providing a conforming alternate version for a non-conforming website?

[patrickhlauke]

Discussed in #2749

[Edit to add TF response comment below.]

^{Originally posted by detlevhfischer October 25, 2022}
Public bodies increasingly fall for promises of accessibility overlay providers that putting an overlay on their sites will make them WCAG 2.1 conformant in a matter of hours. In all cases we know of this is not true, and the promise grossly misleading. Moreover, most overlay tools we have investigated have themselves some or many accessibility barriers in the settings dialog offered to users for customisation of the site. But this is not the topic I would like the WG to clarify here.

This question is this:
In the (unlikely) case that an overlay tool is actually able to successfully remedy all WCAG fails of a site via a specific set of user settings, can this state be considered a conforming alternate version (CAV)?

Looking at the Conformance requirements and the definition of CAV, the situation seems a bit fuzzy.

Section 5.2. Conformance Requirements talks about one specific version ("…or a conforming alternate version is provided") and the definition of conforming alternate versions also talks about "the conforming alternate version" (singular) . One note explicitly qualifiies "One version would need to be fully conformant in order to meet conformance requirement 1."

Where it gets fuzzy is with the last note of the CAV definition, which says:

"Setting user preferences within the content to produce a conforming version is an acceptable mechanism for reaching another version as long as the method used to set the preferences is accessibility supported."

There is no stipulation that just one switch (say, a style switcher typically used to load a more contrasty version of a page) or one link to a CAV is mandated. However, I would argue that without an explicit label or an explicit description in the overlay interface what settings users would need to change to produce the CAV, the mechanism cannot be considered accessibility supported (arguably failing 2.4.6 Headings and Labels).

If this argument is the position the Working Group would adopt, it would follow that a generic overlay without some explicit signposting how to reach the CAV can never be considered a CAV even in the (very unlikely) case that the overlay would be able to remedy all WCAG failures of a site with some combination of settings.

[yatil]

Notes are non-normative. While preferences can be used as a conforming alternate version, going through several versions to find the setting combination that conforms to WCAG is out of reach in my opinion. If overlays could (this is impossible) generate a conforming WCAG version, and would have an accessibility supported way to go to that conforming alternative version, it might be sufficient. But then the question is: Why not provide this WCAG compliant version in the first place instead of taxing disabled people with finding a link to get to the information that they have a right to get to without obstacles.

This theoretical discussion is not really useful. We have better things to do with our time. Don’t get distracted by theoretical possibilities of scamming tools, that is how they make accessibility expertise even more scarce than it is right now, furthering their agenda.

Let’s discuss this when there is a tool that can produce a reliable, 100% conforming alternate version. (Which, again, is impossible.)

[detlevhfischer]

This theoretical discussion is not really useful.

It is not really theoretical for us - public bodies actually ask us whether it is not sufficient to just slap an overlay tool on their site. They don't have the knowledge to see through the fraudulent claims of overlay providers, and have an incentive to believe that a bit of script is enough - it's quicker and cheaper. That's why a decision by the WG would be very useful - we could then point public bodies to it. It would have more authority than any explanation by a single organisation.

[yatil]

It is not really theoretical for us - public bodies actually ask us whether it is not sufficient to just slap an overlay tool on their site.

As overlays cannot (it is, as I wrote above, and repeat here) provide a conforming version, or a conforming alternate version, the answer is NO, YOU CANNOT DO THAT.

I have yet seen one example of an overlay tool actually providing a conforming version of anything. It just does not exist. Unless there is a reliable demonstration with varying content, this whole discussion is theoretical. The scam tools just cannot do it, so they can’t be used for it.

[detlevhfischer]

If overlays could (this is impossible) generate a conforming WCAG version

It is highly unlikely, but theoretically possible - think of a site which fully conforms, with the single exception of 1.4.3 Text Contrast. Activating just one setting in an overlay tool might then make this site conformant. That is why my argument focuses on the accessibility-supported bit - the user would still not know which setting would produce the CAV since it is not labelled as such in the overlay. (And of course I know that in practice, there are nearly always multiple issues which an overlay cannot rectify).

[aardrian]

At the minimum, I would imagine the vendor would need to provide an attestation and actionable contract guarantee that:

• the product conforms to the WCAG definition of "conforming alternate version";
• which includes the "same human language" clause;
• and identifies that the product itself must offer an accessibility supported method of activation/use (meaning it does not break existing AT);
• obviously ensure the site and its product confirm to WCAG at the specified level;
• load from the site itself to avoid tracking protection and ad blockers from killing the third-party script reference; and
• maintain these guarantees for the entirety of the agreement.

When I say "actionable contract guarantee", I am thinking legal protection in case of gaps, refunds for monies spent, enforced timelines to address gaps, etc.

Essentially the kind of stuff we expect from consultants today but tweaked to account for the promises (the legal guarantee since no consultancy today guarantees absolute compliance).

I would also suggest that if the overlay is called from a third-party site, it is too likely to be auto-blocked and therefore cannot (by definition) satisfy the requirement.

If, on top of this, the overlay also actually works then sure, in this theoretical model it could satisfy. But at that point it might be prohibitively expensive or imaginary.

Given the current trend of documented failures to conform to WCAG in the overlay tools themselves, and the inability for them to remediate much of anything on a site, compounded by their reliance on loading from a third-party site, they are not on track to ever hit that mark IMO.

[GreggVan]

(this is impossible)

A good friend and colleague of mine Clayton Lewis (when I said something was impossible or couldn’t be done) reminded me that we should never say "can’t be done" We should only say "we don’t know how to" Things change We learn New things are invented. Gregg

[yatil]

No, there is no way that an overlay can infer the intention of the author of the website. That is just impossible. It can make assumption, but it cannot conjure WCAG compliance out of thin air. Maybe in 50 years. But let’s keep this real and practical.

[GreggVan]

I think it would be great to see a definitive study that answered this question. Like speech synthesis papers in the past though — the most we can say is that "as of the date of this study, there (is or isn’t) any over lay that can make these non-conforming pages conform. There will always be tomorrow - and things may change There will always be some non-conforming site - that has a simple problem that CAN be fixed by the overlay - so overlays always can fix some site somewhere. But since they are being sold and used so extensively - I think it would be good to have someone do a study something like - randomly pick 100 or 200 sites or pages that use overlays - evaluate their conformance to WCAG - publish the results. (For comparison it would be good to also randomly pick 100 or 200 sites or pages that were evaluated and repaired by qualified web accessibility people - and post those results alongside. To show that nothing is perfect but to give an example of the difference between overlays and professionals.) [edit to trim off cruft from email reply to comment]

[GreggVan]

This is no different than what I said. We don’t know how yet. But someday we will. Don’t know when but - someday. PS overlays (at least some) ask the authors for more information in order to complete their repair functions) PPS I am NOT a fan of overlays. Just need to stand on firm ground when evaluating them. [edit to trim off cruft from email reply to comment]

[yatil]

I have looked at dozens of sites with overlays and not one had their accessibility improved in meaningful ways, mostly additional complications and barriers have been introduced. Let’s just assume that unless we have the definitive proof that a general use inaccessible site is made 100% WCAG compliant using an overlay, it is not possible yet.

This discussion is for now, not on an infinite timescale.

[stefanFarnetani]

The question is not about whether overlays can provide a fully WCAG-compliant version at the present (or future) point in time.

In my view, it is a question of whether they cannot do so according to the definition of WCAG.
WCAG requires that "a" compliant alternative version must be achieved. When multiple settings are made in techniques such as overlays, they thereby provide multiple versions (not fully compliant ones).

This question is certainly very sophisticated. Why is this question so important?

If we can determine from the definition of WCAG (Conformance Requirements) that overlays are not a sufficient technique to produce a conforming alternate version, we need not concern ourselves with overlays any more.

This would especially affect testing. Overlays would not have to be considered in testing. Not even in situations like Detlev described in the example (only one success criterion is fulfilled by an overlay).

However, as long as overlays are permitted to deliver a conforming alternative version, we must continue to discuss the state of the art of overlays at regular intervals. In addition, each auditor must check on a case-by-case basis whether the overlays solve anything in the current project.

I do not see sufficient indication in the current wording of WCAG that techniques such as overlays that create multiple versions are not allowed in principle.
If the spirit of the WCAG means that overlays are not sufficient, I think the section "5.2 Conformance Requirements" should be revised.

[yatil]

The question is not about whether overlays can provide a fully WCAG-compliant version at the present (or future) point in time.

In my view, it is a question of whether they cannot do so according to the definition of WCAG.
WCAG requires that "a" compliant alternative version must be achieved. When multiple settings are made in techniques such as overlays, they thereby provide multiple versions (not fully compliant ones).

“A conforming(!sic) alternative version” must fully conform to WCAG. There is no wiggle room there.

The definition of ”conforming alternate version” states clearly, without any room for doubt that it is a:

version that (1) conforms at the designated level

So the alternate has to conform. Full stop.

In theory, sure, if you have a website where all your buttons are links (and this is your only issue), and an overlay swoops in and replaces all links with buttons (for example using ARIA) by the click of a button, then yes, you would have a conforming alternative version.

And that is not what overlays promise and try to do. If they were used to just patch up smaller mishaps, who cares? But they are promising that you add them to a non-conforming website (actually in their fine print they demand your website to conform to WCAG, making this redundant anyway) and through a combination of button presses, you get to a WCAG conforming website. Which, again, none of these things do.

Please use the definitions of WCAG and quote them. Show how WCAG allows a combination of partially or non-conforming versions to constitute a conforming alternate version. It doesn’t.

[awkawk]

I'm surprised by the amount of discussion on a question that seems, at least to me, pretty simple.

Overlays are not currently sufficient, we seem to agree on that, but that isn't the question for me. I hope that AI/ML advance to the point where overlays or other tools which require little or possibly no human intervention can be trained to deliver an improved version. If they can, and that means that all of the criteria in WCAG are met, then there is no reason that they wouldn't qualify as a conforming alternate version.

This is not the current state, but at such time that this is possible, whether six months from now or 50 years from now, the results would need to be evaluated against the SC (or whatever we call them in 50 years!) and a determination about whether the generated version qualifies.

[patrickhlauke]

I think the specific mention of overlay tools here has been a red herring. At a higher level, I think the question is more: if a page uses client-side scripting on top of its "regular" markup that it delivers to the browser to fix, at runtime, accessibility issues, is that acceptable? To which I believe the answer is yes, of course (provided that the scripting works correctly, which is a separate discussion, and the one where the whole "overlays can't actually automagically fix accessibility" discussion then goes on). Particularly since these days, the line between "server-sent page" and "processed client-side" is so utterly blurred with things like React etc.

[yatil]

I think the mention of overlays is important because they currently only (try to) fix individual issues – contrast here, text size there, adding weird ARIA somewhere else. That means to get from a non-conforming version to a conforming version, you would need to go through multiple non-conforming versions. That is not covered by the definition. If a tool provides a conforming version with one click (and that click is accessible as defined), then yes, it would be sufficient. But there cannot be layers of inaccessible versions.