To answer the questions in your blog post :

“My research is to see what authors are doing and whether hints like
<td><b> are reliable enough for Hypertext Markup Language 5 to specify an algorithm which User Agents would be expected to implement. ”

Currently browsers do display this as bold text inside a table cell. However, screen readers don’t recognise styles. If this were to be used instead of <th>, it may be years beforea ll screen readers treated this as a header and all, or most, screen reader users had upgraded to the version that did support it. I’m also not sure how a screen reader would be able to tell the difference between an HTML 4 page using this just for styling and an HTML 5 page using it as headers.

Your other question:

“While I’m here, do you find that treating the first row as column headers and the first column as row headers works a lot of the time? If it does, that might help figure out tables which don’t even use <td><b>.”

Most of the time yes, the topmost row and left-hand column are the logical headers, but this is the point of this article. What happens when they aren’t? We can’t really expect the sighted world to change it’s natural order, so that the left-hand column always held the logical headers.

Hope I haven’t misunderstood the points you were making Ben, keep up the good work.

I don’t think user agents could reliably choose which cells contain headers though. It’s a decision that needs to be based on knowledge, which is why I suggest that authors refer to the person who compiled the table, if the key information isn’t obvious.

Would the user agent “make decisions” based on the number of characters in the data cell, or position, for instance? These could be programmatically determined, but could so easily identify false “key” data and cause more confusion than they cure.

Would it be useful if UAs could figure out which cells are headers even if they aren’t marked up with <th>? Much like yourselves, the majority of tables I come across don’t use it (yet).