Given Monte Carlo Simulations are the base of what Range does for customers projections, making changes could be high-stakes since a tiny rounding error could throw off a retirement projection. How did you guys set up your testing to catch subtle issues?
That was definitely something we were concerned about. We ran the new logic against all households and compared that to the output of the existing logic. We used this iterative loop until the difference was 0 which gave us confidence in the implementation.
Given Monte Carlo Simulations are the base of what Range does for customers projections, making changes could be high-stakes since a tiny rounding error could throw off a retirement projection. How did you guys set up your testing to catch subtle issues?
That was definitely something we were concerned about. We ran the new logic against all households and compared that to the output of the existing logic. We used this iterative loop until the difference was 0 which gave us confidence in the implementation.