I have been thinking about figures. Often when coded correctly, a screen reader will speak fairly much the same information 3 to 4 times on a figure.
1. When the figure is encountered the figcaption might be a label for the figure so will be spoken when visiting the figure.
2. The omg will have an alt tag on New York Times and Washington Post I have noticed that the alt text of the image is nearly identical to the figcaption.
3. The figcaption is the visible text below the picture and descries the picture for the sighted.
Maybe 4. Then the end of the figure is encountered. I’m not sure if NVDA and JAWS will also read the label at this point.

Some possible improvements:
1. Ignore alt text in images if they are in a figure that is labeled with the same content as the figcaption.
2. Don’t read the figure label when the start/end tags are read if the label of the figure is the figcaption.
3. If there are options to not read start/end tags for articles and regions then allow figures to be treated the same way. (This might even mean a different behavior for Red all vs navigating through.)

Jonathan