It's a little hugged to death at the moment, but https://covid19risk.biosci.gatech.edu/ is a useful tool to show relatives who are thinking about social gatherings for the holidays.
Lead dev here... Huh, didn't think I'd ever see my work show up on HN. Sadly getting an .edu institution to fork over the money for better scaling is hard... Even when it gets them unlimited free press.
Just a reminder, that tool likely underestimates the risk of covid. Seroprevalence data shows undercounting of up to 24 times (rather than the 5 or 10 this tool shows.)
Also, this tool assumes you are only seeing people from your county.
And humans are terrible at assessing compound risk. If the tool shows a 5% risk from a gathering, people will likely attend multiple events. ("Oh, it's only 5% to go to Thanksgiving. And 5% to go to Friendsgiving. etc.") When you consider multiple events, risk doesn't stay constant.
So while that tool is interesting, don't look at it as a mechanism to say, "Yeah, it's probably ok to have a gathering."
While you're not wrong that we might be under-estimating risk, I'll note that we do take into consideration seroprevalence in our calculations (so called, ascertainment bias). We provide estimates for two cases; 5x and 10x under-counting bias.
And you are right, that risk is additive (to a degree) if you go do multiple parties. And people are terrible at doing this sort of calculation. Our messaging, in our manuscript/press/social media has always been to limit gatherings of any size, and mask and gather outside as much as possible
It's also worth noting that this tool shows the risk of at least one other person in attendance having covid before going, not the risk of contracting covid at the event.