Improvements to screen reader report from graph

[kmcnaught]

Some comments, to discuss improvements to default behaviour, or surfacing multiple options

Currently the live-region reports "Estimated action: "{action}""

Questions/suggestions:

• Is this repeated based on a timer, or just repeats once it's finished?
• Consider if it should report based on transitions, so it's more time sensitive? Obviously with some kind of debounce/hysteresis to avoid too-rapid change.
• Consider whether the report "certainty for {action}" should also give continuous updates on a single class? I can see situations where this would be helpful.
• How is it expected to navigate to the graph for updates? In my experience playing with NVDA:
  (a) the updates happened automatically when I started testing the model
  (b) I then lost focus and struggled to get back there
  (c) I now (with visual help) can get there via arrow key navigation in NVDA
• Have I missed an expected way to get there?
• Consider adding a region to allow quick navigation

[kmcnaught]

Add:

• Consider removing "Estimated action" after first announcement for less verbosity / less latency on new information, e.g. "estimated action wave... shake ... shake ... nothing ... shake"

[microbit-robert]

The estimated action and certainty values update every 250ms. The estimated action live-region is announced when it changes from one action to another, e.g., still to shake. What actually happens in practice is a transition from still, to unknown and then to shake. Therefore, the announcement can get out of sync quite quickly. We can look at debouncing the announcement.

I'm worried about removing "estimated action" after the first announcement. If you don't move your micro:bit for a while, then accidentally knock it, it's going to say "shake" without any further context.

We'd need to think about how adding certainty for {action} is going to work together with what we already have. So I don't think we should rush to add this.

I haven't managed to lose focus. It will always read out the estimated action when it changes, so I'd like to know if this can be reproduced if you can get it into the same state?

[microbit-robert]

We discussed the possibility of only focussed regions reading out either the estimated action or the confidence level of the focussed action row. It might be nice for the user to opt-in to these annoucements in this way rather than having estimated action always announced.

[microbit-matt-hillsdon]

We discussed the possibility of only focussed regions reading out either the estimated action or the confidence level of the focussed action row. It might be nice for the user to opt-in to these annoucements in this way rather than having estimated action always announced.

If we do this then we should also consider keyboard shortcuts/access to focus the region.