As you say, sars-cov2 viruses come with all sorts of epitopes that could conceivably generate immune responses, and the sense I get is that with real infections

As you say, sars-cov2 viruses come with all sorts of epitopes that could conceivably generate immune responses, and the sense I get is that with real infections, there's an element of chance as to which ones generate responses, and they're also not all created equal in terms of which ones tend to be conserved vs. vary a lot among strains. The choice of the spike protein for the vaccine was specifically to generate responses to epitopes that are thought to be highly conserved (though not perfectly, given e.g. B1351), whereas with natural infection you might generate a response that successfully clears your current infection but might be less protective against the next version you encounter. (Though at 84%, maybe that effect is weaker than some speculated it might be?)