W3C

– DRAFT –
ARIA and Assistive Technologies Community Group Weekly

07 May 2025

Attendees

Present
Carmen, ChrisCuellar, howard-e, IsaDC, james, jugglinmike, Louis, mmoss
Regrets
-
Chair
-
Scribe
jugglinmike

Meeting minutes

<mfairchild> regrets for today

Review agenda and next meeting dates

https://github.com/w3c/aria-at/wiki/May-7%2C-2025-Agenda

Matt_King: Requests for changes to agenda?

Matt_King: Hearing none, we'll stick with the agenda as planned

Matt_King: We may want to cancel next week's meeting

Matt_King: The way the outlook appears right now, we won't have new stuff for the test queue this week. That will all be available the following week

Matt_King: So we might be able to save people some time by not having a meeting next week

Matt_King: Does anybody anticipate anything that I'm not anticipating that suggests we should meet next week?

IsaDC: Not from our side, I don't think

Matt_King: Okay. I'll make a decision based on how things are looking after Monday's meeting with PAC

Matt_King: And also based on anything else that may turn up

Matt_King: Following Community Group Meeting would be: Wednesday May 21

Matt_King: Next AT Driver Subgroup meeting: Monday May 12

Current status

Matt_King: We have three things in draft review, but they're all held up by one thing or another

Matt_King: We'll talk about radio group when we talk about arrow keys and NVDA later on in this meeting

Matt_King: Vertical temperature slider has incoming changes which may be done by the end of the month

Matt_King: As for the disclosure navigation: we could re-evaluate and push it on through, but it's kind of a waste of time when we know at least (at a minimum) are coming to JAWS

Matt_King: We've learned that those changes in JAWS won't be available in the May release, but instead in the July release

Matt_King: As for what's coming next

Matt_King: James and IsaDC and I keep running into a variety of problems which make some options untenable at the current moment

Matt_King: IsaDC is planning to get the work on according done by the end of next week

Matt_King: The next three that I have listed here were all scheduled to get done in April or May, but they all ran into strange problems

Matt_King: There isn't clarity in the HTML AAM about how screen readers should handle lists made with dl/dt elements.

Matt_King: But it's not really necessary to use that specific structure in the APG, so James is going to raise an issue in the next few days about that

Matt_King: The image description one is even messier because we actually don't know what screen reader expectations ought to be for figures (especially for labeling figures)

Matt_King: I have an action item to file an issue against HTML AAM or maybe even accname

Matt_King: But the uncertainty there may delay this work for a really long time

Matt_King: In summary, those are the kinds of technicalities which are delaying the upcoming work

Matt_King: But it will be accordion next, and the spreadsheet shows what will follow that (I believe it's going to be the spin button)

Re-run of JAWS for color viewer slider

Matt_King: I re-tested this with not the exact same build that Hadi was using, but with a beta build that followed Hadi's. Brett assured me that there were no relevant changes between these releases. My results matched Joe's results and not Hadi's results

Matt_King: I have a feeling this was just a copy-and-paste error on Hadi's part

Matt_King: Can you e-mail Hadi, IsaDC?

IsaDC: Sure

IsaDC: Were they running the same version?

Matt_King: No, Joe was running the March version and Hadi was running the April version. I believe that's what we determined from the meeting minutes, anyway

Matt_King: I think this is going to be a fast fix. Once that's done, we can just push this

Arrow keys in NVDA test plans

Matt_King: There is a setting in NVDA that is enabled by default. When things are visually on the same line, the arrow key will put a whole bunch of elements (in this case, 5 radio buttons) on the same line, so that the arrow key doesn't semantically separate them; it just reads them all at once

Matt_King: That's different from how both JAWS and VoiceOver work by default

Matt_King: This has consequences for the rating radio group

Matt_King: I went back and looked at what we did for the other two radio groups, and it's quite interesting that we did use "insert + up arrow" for requesting information, but we left the arrow keys out for the "browse mode" testing for navigating from one radio group button to another

Matt_King: I'm not sure why that is

Matt_King: Then I started thinking about this and kind of wondering: would the community group be aligned with expecting NVDA to by default NOT combining these things? This is about conveying semantics and not about conveying visual information

Matt_King: It doesn't feel like an inconsistency that we want to support

James: I think the screen layout setting in NVDA is not a faithful representation of what's on the screen. There are cases where some things are on the same line, but NVDA will not convey them as such. I think there are inconsistencies and limitations from what they can get from the browser

James: The setting is intended to convey things on the same line

Matt_King: I'm saying that for that to be the default behavior--so it gives a different semantic behavior by default

Matt_King: In some cases, the arrow key can move you to a specific radio button in a radio group, but in other cases, it cannot

Matt_King: Both cases are semantically equivalent; the difference between those cases depends only on the CSS

Matt_King: It either means we have to tell people (e.g. testers), "never use the arrow key". But real-life user do!

Matt_King: ...Or, we could possibly say to NVAccess: "fix the interoperability bugs that we're finding by changing the default value of this setting""

James: The aim of the setting is not to change the semantics but to change how the controls are conveyed

IsaDC: It's for the navigation--how you move through the buttons. You can't do it with the arrow keys

James: I think the default is terrible, and so did all the folks who responded to my informal Twitter poll

James: That's just my opinion, though

James: I can't pick out all the information--that's why I dislike the setting

James: I agree from a cognitive perspective. But that isn't semantic; it's just a subjective opinions

IsaDC: This concerns the tests for navigate to next and previous radio button, right?

Matt_King: Yes

Matt_King: In one case, you get one button, and in another case, you get a bunch of buttons, and so the way the command is behaving, by default, is giving you a completely different amount of--I guess you can distinguish, if you listen closely, you may be able to differentiate the radio button labels

James: We have discussed a number of cases where there is a fine line between objectivity and opinion. And we agree that we're not trying to change the screen readers' decisions about how they present information, as long as they convey that information.

IsaDC: What is the output we seek from this issue?

Matt_King: Essentially, we don't know when it's okay to test with the arrow keys and when it's not

Matt_King: What we're currently telling the world is that the commands you can use for this pattern vary based on visual presentation. So the commands end up based not on the pattern but on the visual presentation of the pattern

James: In that informal poll I conducted, respondents offered a bunch of work arounds

James: The APG could make a vertical radio button example

James: It feels like we're overstepping the mark, even though we all agree that the default behavior is not good

Matt_King: That's exactly my question: would we be overstepping the mark on this case?

Matt_King: And I'm hearing that you think it's better to not test with the arrow keys for NVDA

IsaDC: Do we have other exceptions like this?

IsaDC: For other screen readers, that is

Matt_King: None come to mind

James: For JAWS, we almost always add an extra arrow key press because it adds inline "fake" text

Matt_King: We do that for VoiceOver, as well. NVDA is the outlier in that case

IsaDC: I think testing the arrow keys would bring more issues

James: If we test this and fail NVDA, I think we're going to get push back from NVAccess or elsewhere that essentially, "it didn't fail but it just didn't act the way you wanted"

Louis: I agree with James

James: It became apparent from NVAccess (because Jamie Teh replied) that they don't have access to data for how users change settings

James: I do think that a company like Vispero will think about this a lot more--they have the resources to do so, and they have more avenues for structured feedback, etc

James: I think that's a shame because whenever I've tried to get people using NVDA who are more comfortable with JAWS, there are defaults which trip them up

James: I think, realistically, NVAccess will just tell us, "it's only one keystroke for users to turn this off"

Matt_King: I'm hearing that it is an overreach for us to weigh in on this particular behavior

Matt_King: I agree that ARIA-AT isn't always the way to push for change

Matt_King: But we should still be consistent within ARIA-AT

Matt_King: We tested with "insert + up arrow" on another test plan, but that's only due to luck

Matt_King: I'm wondering: for consistency within the tests, even if "insert + up arrow" happens to work in some tests, should we include that command?

Matt_King: Or should we use the same commands across all radio group test plans?

James: All three radio group test plans in the APG put the radio groups on the same line

Matt_King: I just ran through them last night, and the pizza crust ones, for example, were putting them on separate lines

James: The best way to test NVDA with defaults is to make a portable version

Matt_King: Henceforth, I will always use a portable NVDA. Is that really the fastest way?

James: You can also rename your NVDA folder in your AppData folder and restart it

Matt_King: How did people get the "insert + up arrow" results that they got for radio button activedescendent?

Matt_King: The output that people recorded was not the options. Does that mean they weren't using the defaults?

IsaDC: I was using the dafaults

Matt_King: Oh, but that was bot output!

IsaDC: I always double-check the bot output

Matt_King: I'm questioning all the NVDA research I did last night

James: I wonder if they act differently in the APG than they do in ARIA-AT

IsaDC: They shouldn't

James: they all work the same for me with the absolute defaults (I haven't even changed the synthesizer). Even in the APG

Matt_King: Then our reports are wrong

James: the only thing I can see changing that is if the browser window was too small

Matt_King: Sometimes the setup script doesn't work correctly if the window is too small

IsaDC: My concern in that case is for the other two

James: I can't make those appear on separately lines by zooming in, either

Matt_King: I think we're going to remove "up", "down", and "insert+up" from all these test plans

Matt_King: But now, I'm very concerned about how we got the results that we got

Matt_King: My only theory is that the bot reported incorrect responses and that no one corrected it

James: It is possible that the bot was in the wrong mode

Matt_King: Yes, because focus mode would change the "insert + up" behavior

Matt_King: But I was getting blank responses when I was in focus mode

Matt_King: Okay, this turned out to be much more complicated than I expected!

James: I'll share a link to the issue I filed against NVDA

<ChrisCuellar> James: nvaccess/nvda#15159

Matt_King: Do we re-run the test plans first? Or do we change them, first?

Matt_King: Let's go with the active descendant. Add the current version of the test plan to the queue, run it with a bot, and let's have a human go through it to determine if it is giving correct output. Let's make sure that whatever human reviews it is truly using defaults (allowing only changes to speech rate change)

Matt_King: Make sure we're using the latest NVDA with defaults

jugglinmike: The NVDA bot is using version 2024.4.1

jugglinmike: https://github.com/bocoup/aria-at-automation-nvda-builds/releases/tag/2024.4.1

IsaDC: I am using 2024.4.2

Matt_King: I don't know if or how much that will matter

Matt_King: We'll run it with the bot and just see what it does

Matt_King: this could also be due to a mode problem as James suggested earlier

Matt_King: We can dig in if we see a discrepancy

Arrow keys in NVDA test plans

Matt_King: We're just about out of time

ChrisCuellar: It's okay, this isn't urgent

ChrisCuellar: We are running the tests in automation over and over again to detect ways that the automation system is inconsistent

ChrisCuellar: In that process, we identified an inconsistency that we could reproduce manually

ChrisCuellar: And it's not clear how we should handle that

Matt_King: Oh, that's interesting

Matt_King: Let's plan to talk about it next time

ChrisCuellar: Sounds good!

Minutes manually created (not a transcript), formatted by scribe.perl version 244 (Thu Feb 27 01:23:09 2025 UTC).

Diagnostics

Maybe present: Matt_King

All speakers: ChrisCuellar, IsaDC, James, jugglinmike, Louis, Matt_King

Active on IRC: ChrisCuellar, howard-e, jugglinmike, Matt_King, mfairchild, mmoss