However it has the benefit of only resorting to reduce, a function that can do anything, to perform its textbook application, hence better conveying intent. Maintanability saves more money than micro-optimization.

Agreed!

for how much cleaner it is, it's only marginially slower than my solution; Only once we get to a test size of 2,500,000 data points does my solution actually start to be "noticiably faster" (180ms vs 320ms)

but how fast ...

nice and tidy

cool approach, nice catch that you can tell which sequence you'll get by looking at 3 values and than just extend out from there

seems easy for a 5kyu

For some reason this exceeds max call stack once the input gets really big

solid solution. Nice work

you're a pro. The way you compare the directions is the best I've seen of all the solutions. Also solved in O(n)

nice type safety

yup, I ran into the same issue