The entire premise of the app is that P[transmission] = F(distance, duration). I think this is a naive assumption and just the two input parameters have nearly zero predictive power, i.e. the variance of the estimated probability would make any further analysis useless. I'd also argue that the only case where it would be useful is tracing a zombie-level virus when there's willingness to lock down with military a 5 sigma radius around the probable infection. Otherwise it's just a tool for marketing and spying.