Resolving to Measuring the Shelter Effect: A Regression-Discontinuity Audit of a Bus-Stop Boarding-Count Cutoff