W3C

– DRAFT –
ARIA and Assistive Technologies Community Group Weekly Teleconference

24 July 2025

Attendees

Present
Carmen, ChrisCuellar, howard-e, IsaDC, James, Joe_Humbert, jugglinmike, louis, Matt_King, mfairchild
Regrets
-
Chair
-
Scribe
jugglinmike, Carmen

Meeting minutes

Review agenda and next meeting dates

https://github.com/w3c/aria-at/wiki/July-24%2C-2025-Agenda

Matt_King: Requests for changes to agenda?

Matt_King: Next CG meeting: Wednesday July 30

Matt_King: Next AT Driver Subgroup meeting: Monday August 11

Current status

Matt_King: we have accordion in the words and three other plans in "draft review"

IsaDC: I opened a pull request for the vertical temperature slider

Matt_King: We're going to be ready next week, then--for people to take a look

IsaDC: We need to mark it as draft

IsaDC: ...just as soon as it appears in the app

Matt_King: I meant to have the "radio" test plan ready for today, but it's not. Hopefully it will be next week

Matt_King: IsaDC and I have been working on tabs

IsaDC: There are two more tests for me to modify, then it will be ready for your review, Matt_King. Expect that by tomorrow

Re-run JAWS report for color viewer slider

Matt_King: We still have two conflicts

IsaDC: I was working on getting those conflicts resolved

Matt_King: There were originally conflicts related to Hadi's results, and those got resolved

IsaDC: I think I will retest for the remaining conflicts because there was a conflict between Windows 10 and 11, so I need to re-run those two with the latest JAWS version

IsaDC: I can do that after this meeting

Matt_King: Looking at the conflicts that are in here--I don't know if this was related to Windows 10 versus Windows 11.

Matt_King: Joe_Humbert was getting the min/max announcement while Hadi was not. It was just for one command

IsaDC: Yes, we retested this during a meeting, and it was flaky

louis: Weren't we going to wait for the July JAWS update before we re-tested some commands?

Matt_King: I think that was for disclosure, but that makes sense here, too

IsaDC: I could try with the JAWS bot because the JAWS bot is available on the staging server

Matt_King: I think we should proceed by editing what we have here instead of starting over

Matt_King: Hadi was using the May update, and Joe_Humbert was using the March update. Neither one of these was with the July release!

Matt_King: I think IsaDC should edit Hadi's results using the July build

Joe_Humbert: I am currently using the version that Hadi was testing on

Matt_King: can you update to the July build, instead?

Joe_Humbert: I can try!

Matt_King: That would be really good--if you could re-do that one test and use the July update

Joe_Humbert: It looks like the latest public release is from July, so yes, I can run the July build

Joe_Humbert: A couple weeks ago, I posted a question: would it be better if I got an actual PC to do testing (so that I'm not using Windows in a virtual machine that is running on old macOS architecture)

Matt_King: That shouldn't make a difference

Run of accordion test plan

Matt_King: there are three issues--2 related to NVDA conflicts and 1 related to JAWS

louis: We can defer discussion on the JAWS conflicts because Joe_Humbert is going to update to the new release and re-test

Joe_Humbert: I'll make sure I'm using the latest JAWS and the latest Chrome. I'm using Windows 10, though

louis: And I'm using Windows 11

Joe_Humbert: Hopefully that doesn't impact our results, but it could

Issue #1270: "Navigate backwards to an expanded accordion header" (Accordion, Test 2

github: w3c/aria-at#1270

louis: I got strange output where it says "heading" but it doesn't say anything about the heading level

Matt_King: I was trying this myself earlier, and I think I was not getting the same result as you...

Matt_King: Joe_Humbert, your output is almost perfect--it's exactly what NVDA should do. We know for sure you're using the correct version of NVDA

Matt_King: For the first time, I took James' suggestion and tested with an NVDA portable copy. The only thing I don't like about that approach is that it takes over a minute to start

Joe_Humbert: I've been using a portable copy, and I always reset the defaults when I'm testing

James: The recommendation was that people make a portable copy that they don't use so that they don't need to reset it

Joe_Humbert: Well, I just download NVDA and use it directly, immediately

James: Yes, I think the recommendation is more useful for people who are using the screen reader under test in their daily lives

Matt_King: I think we're aligned, then--everyone is testing the way they need to be testing. But the two testers are still getting quite different results

Matt_King: Joe_Humbert, is there a chance of transcription error, since you aren't using the speech history?

Joe_Humbert: No. I am very careful and go back and double-check

James: Louis reports that the screen reader says "billing address" twice, but that phrase doesn't appear in the reported output

louis: That's a typo in the error description. It's actually a different piece of information which is duplicated

James: Heading levels never spoken by NVDA when it's in focus mode

James: The fact that the heading level appears in the output reported by Joe_Humbert suggests that NVDA was in browse mode at the time--perhaps it switched to browse mode

James: When I switch to browse mode, it includes a "boop" sound at the end to suggest that it is in browse mode

James: I think the issue is with our instruction to testers to begin by ensuring they are in focus mode. If you switch to focus mode yourself and then press "shift+tab", then it will not switch

James: [demonstrates the behavior of NVDA across different contexts]

Matt_King: This is an interesting thing. Because this is a navigation test, the way it actually switches is what we expect

James: When we're testing specifically, "navigate backwards after switching to focus mode"

James: And there is no way to check your current mode in NVDA. When instructed to verify the mode, testers much toggle back and forth, and that changes NVDA's subsequent behavior

Matt_King: So if you follow our instructions to a "T" in this case, you get the wrong results

Matt_King: But if you test for focus mode in a way that doesn't interfere with the test...

Matt_King: We say "navigate backwards in focus mode". That is the test

Matt_King: Do we have a "mode switch" assertion?

IsaDC: No, because this is a switch in reverse

Matt_King: Ah, that's right, we intentionally don't test for cases like this

Matt_King: Loius, the way Joe_Humbert did it is correct. I think we need to change your results

Matt_King: I don't know if we're going to do something to our testing instructions as a result of this discussion

James: We could make an add-on for NVDA that reports the current mode

jugglinmike: This also impacts automation. In both JAWS and NVDA, we likewise don't have a programmatic way to detect the current mode, so we toggle mode once, and potentially toggle a second time

Matt_King: This may indeed be a relevant problem for automation. Let's add it to the agenda of the next "automation subgroup" meeting

TPAC

Matt_King: We're short on time, so we'll talk about this next week

Output normalization

github: w3c/aria-at-app#1462

jugglinmike we're commiting to automation in the recent weeks and we have expanded it's power and extent in which we are relying on it

It has surfaced some new issues that we need to walk through. One of them is normalization

mattl Do we store a normalized state or do we normalize every time we want to compare ?

And never store the normalized form

jugglinmike we never store the normalized form

I do make the distinction in my issue. We have to discuss when we normalize and what we normalize

The two time we consider normalization is when we display the output and when you compare the output

Maybe we can save the concept of saving as an implementation detail

Folks will interact with this when they are reading it and when they are using it for comparaison

Is it easier to talk about what normalization should happen and then when. I thought they are linked

mattl A part of me thinks we should never display normalized outputs

@james We don't want to see (didnt catch it)

mattl I thought the bot wasn't doing what is was supposed to do

@james Mike has a list on his issue which is asserted that there are discrepency that don't have any bering on the output for the end user. I am not sure I agree. Punction is very important for fluctuation

mattl I agree, specially on VO

@james Unicode group separator would never make it to the speach output so that is quite different to punctuation which does make it to the user

mattl I agree with what you said. The unicode separators fall in their own class

I don't think they show up in screen history. We only got it from the the JAWS bot

jugglinmike that is right, only with JAWS bot

There are some other one too. It's good to know that the punctuation can have bearing. I should ask: How easy is it for you as a tester to consistently report that information. Do you know if there should be a dash betwen ctrl and alt when you listen to it

@james I guess it doesn't matter because that is why we are copy pasting

mattl VO captures all punctuations

jugglinmike I like that but I have to say the two reluctantcies I have is that we are commiting to those tools being necessary part of the manual testing

mattl I guess sighted testeers will se them

@james I think the daily AT users have an intuition for the punctuation but there is a subtle difference between a period and a colon or a semi colon and a coma

From that point of view the answer is that a human intuition about punctuation is much less accurate

mattl So what you are saying is that we def want to capture the speech output faithfully. We want the actual string coming from the AT

jugglinmike That represents is a difference or the way we have talked historically. Maybe we would need to do some documentation. But it is helpful to know our expectations from human testers and the machine

@james you mention empty spaces. IF peole use the speech histry for NVDA they will have two space character between chunks of outputs where chunks could be undefined. But usually means if it says Heading Level 3 there will be two spaces. And so those in the speech by NVDA are separate. When it is rended to text it put two spaces so that you have a

better chance to determine where those gaps were

jugglinmike In terms of the algorithm for normalization we might want to normalize more than two spaces. it might be good to normalize it because we wouldn't want to invalid our algorithm.

@james If a string on a page ended with a space, the speech output will include a space and not say a space

There are some differences

mattl this is why the spaces conversation was important. In JAWS new lines helps understand what was part of the output or the screen that is read or what JAWS is telling you about the test in the screen. So this is where maintaining the original output can be helpful

But when it comes to comaring historicaly because NVDA and JAWS and VO they might make changes that doesn't affect what is spoken but do affect the output. Maybe they use a line instead of two spaces. So I think that when we are doing comparaisons that we would always compare normalized values so any double white space becomes 1 white space and

always the same regardless of the new line

@james I don't think it will contain two outpts. That is an implementation of the speech details. What gets sent to the synthetizer is a series of commands. It s a whole thing

I think I agree that what humans are seeing should reflect what was gathered and internally when you compare that is an implementation output and a lot more normalization is appropriate

Carmen Next step is to outline normalization in the issue and reach out again

We can also send an email to everyone in the CG

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

Diagnostics

All speakers: IsaDC, James, Joe_Humbert, jugglinmike, louis, Matt_King

Active on IRC: Carmen, ChrisCuellar, jugglinmike