it's currently C# 10.0 from .NET 6.0, and I can use destructuring assignment just fine. Closing, as the C# version got updated

Kata is now retired.

You are the best!

My first implementation passed all tests, eventhough it evaluated
"1 + 2 * 3 * 4" to 28 (the correct answer is 25)

my bad, error was in my code
please remove/resolve this issue

Having a closer look at both the Random tests that pass and fail, I have a suspicion this is caused by the '/' symbol - which is shared by all three modes

I have failing tests when I attempt this in C# (in the Random and SymbolsEdge sets), which I can't explain.

For example, my solution fails on the following inputs:
"2>2 ZmU4sm7v.#56& e#M@ ¿ /+T," (Random test says: Expected value 216, my solution gives 214)
"Gk0_ @HN?/" (Random test says: Expected value 62, my solution gives 61)

I've tried the latter one manually, also arriving at 61. Here's the output:
from: (0, 0)
to: G (1, 1), mode: 2
takes: 1 shifts, totalling 6 presses
total: 6
from: (1, 1)
to: k (2, 0), mode: 1
takes: 2 shifts, totalling 9 presses
total: 15
from: (2, 0)
to: 0 (3, 7), mode: 1
takes: 0 shifts, totalling 3 presses
total: 18
from: (3, 7)
to: _ (4, 6), mode: 1
takes: 0 shifts, totalling 3 presses
total: 21
from: (4, 6)
to: (5, 1), mode: 1
takes: 0 shifts, totalling 5 presses
total: 26
from: (5, 1)
to: @ (3, 6), mode: 1
takes: 0 shifts, totalling 6 presses
total: 32
from: (3, 6)
to: H (1, 2), mode: 2
takes: 1 shifts, totalling 10 presses
total: 42
from: (1, 2)
to: N (2, 3), mode: 2
takes: 0 shifts, totalling 3 presses
total: 45
from: (2, 3)
to: ? (0, 2), mode: 3
takes: 1 shifts, totalling 11 presses
total: 56
from: (0, 2)
to: / (4, 8), mode: 3
takes: 0 shifts, totalling 5 presses
total: 61

I don't look ahead in the string to optimize for shift.

There's more to a Best Practice than asymptotic performance, e.g. readability and expressiveness.

Out of curiosity, what do you think the asymptotic performance of this solution is?