I'll take a look at this tonight or tomorrow. Thanks for reporting the issue!

If they're trying to represent probabilities then the cap should actually be 1-(1/e) where e is Euler's constant (2.71828...) or approximately 63%. That's still not really an upper limit, just a useful benchmark for these extreme cases. For instance 2 million microcovids would be roughly 86%. It gets asymptotically close to 100% but not quite as fast as one might think. For instance a 99% probability corresponds to a whopping 4.6 million microcovids.

If they aren't trying to translate their sums into probabilities there's no need for a cap. There's no reason not to count how many times you roll a million sided die, without bothering with a cap, as long as you don't interpret that as something it isn't.

They certainly shouldn't be claiming that 1,000,000 microcovids represents 100% probability, because it doesn't.

For very large values of n and comparatively small values of m (1-(1/n))^m is close to 1-(m/n) (binomial expansion) but this starts to fail as m gets close to n. I suppose a quick fix would be to approximate by n*(1-e^(-m/n)) where n= a million and m=number of microcovids. It's roughly equivalent to m for small m. Or at least start adding higher order terms to the Taylor series!

I think this is actually a bit instructive as far as mental health is concerned. By now we've all met people who got Covid but insist they've been doing "everything right". Those people no doubt include a fair number who use this site. The thing is, if a bunch of people spread out a million microcovids over the same period of time - be it a day, month, year, whatever - approximately 63% - that is, 1-(1/e) - will get Covid. A lucky 37% (1/e) won't. Those are exactly the kind of numbers that might lead people to infer a lot more agency than actually exists. Or, at the other extreme, infer the existence of a malicious demon who doesn't care how careful you're being.

All that being said, this issue strikes me as a poster child for mission creep. Why anyone would care about probabilities exceeding 50% is beyond me, unless they're actively trying to expose themselves (don't do that). This tool was designed strictly for weekly risk budgets with an already cautious audience, calibrated to the virus's incubation period, and the kinds of distortions you get with large numbers don't have a lot of relevance (except for the mental health benefit described above). Whatever you did several weeks ago has very little bearing on what you get to do this week.

FYI, I'm calling 1-(1/e) the Universal Unfairness Constant.