I think there's actually something wrong with the calculation.
For instance, it doesn't appear to control for sample size by Age group and Height (e.g., weighted average). Meaning, it's sensitive to sample size (i.e., sampling bias). So, just having more/less data points for 20-30 year olds than 30-50 year olds will skew the outcome (and, surprise, it does).
For example, if I expand the Age window from 30-50 to 20-50 (holding other setting constant; Weight, Race = "Any", Income = 80K, Height = 5'11") I would expect my success rate to increase, as I am increasing the pool of applicants. However, the tool shows a decrease (30-50 = 9.2%; 20-50 = 6.2%). If I change the Height to 6", we again see a reversal (30-50 = 5.5%; 20-50 = 3.9%).
Guess it's hard to find a guy who knows statistics too.