I think we’re in agreement, just that you understood what I wrote a bit differently from how I intended. By “described in the visible text” I didn’t mean “expounded at length”. I’m talking most commonly about the sort of captions newspapers printed with images, typically 5–30 words; but also about things in the text that refer to the figure, like “here are the benchmark results:” and “we see that X is 57% faster than Y”.

There’s no good way of exposing lengthy information about a chart to screen readers. Don’t try doing it via alt, it’s a bad experience; and don’t try doing it by screen-reader-only text, that just gets confusing. The longdesc attribute was an attempt to do this, but it failed to gain traction among users or assistance tech. Consensus now is that you should instead include in the caption a link to a page that focuses just on the chart, which can then include more detail for the screen readers.