I'm looking for help understanding these walk-time results:

All the points on the screenshot were in the input. They have been shared to AGOL from Pro in a web map. There are <1000 points in the dataset.
Time bands had breaks at 2, 4, 6, 8, 10, 15, 20 minutes.
Towards the centre of the map are two points in a dark purple area, next to a pale purple area with no points in it. Other points show as expected, in pale purple (short walk) areas.
Any suggestions?
We tried using just a subset of points and the problem remained. It is consistent between analyses so I suspect something to do with point location.