Dead-reckoning arrival never fires if the last step counts to the exact metre: add a tolerance
A step-counter position engine that finished the final route step only when walked distance reached its length never declared arrival in simulation: the simulated walker covered 80.5 m of an 80.3 m route, yet the engine counted 2.4 m of the last 3.0 m. Each RFID snap resets the step origin to the current sample and step counts are quantized, so a little distance is lost per reset, and real stride lengths vary by several percent anyway. A walker who stops 0.6 m short would never be marked arrived. Count the final step as done within about 1 m of its end (leave other steps strict, they catch up as the walker keeps moving) and test both sides of the tolerance. Running a simulator over every new route found this; unit tests on one route did not.