This is a standard application of Dijkstra's algorithm, which is a duplicate to existing katas.

I get a message:

There was a server issue while sending the request. Please try again

every time I attempt (and can't submit with all tests passed).

Second sample test case has the wrong expected in the sample tests (2.282 instead of 2.828)

Same as Galaxy Getaway, hardcoded massive cases are NOT random cases, and it's a waste of bandwidth.
Ofc you did not code those cases by hand, right?
Then just replace them with your random cases generator

Randomize portal names IS NOT random test
You need to generate randomized map on the fly
Also you hardcoded massive data that force everyone to submit 1.08+MB data while attempting, making it very hard to successfully get code executed