The aim is to have the most cards at the end. Not take the most of them each turn.
Sometimes it's better to just take one.

Remember the instructions : Bob also tries to play optimally.

In your case, Bob could take 9 cards by playing it smarter so Alice would be able to only take 4 cards.

n = 69 : expected 30n to equal 8n - is too little number
n = 85853 : expected 42906n to equal 26n - is too little number
For n = 428708124794396147 : expected 107177031198598953n to equal 88n - is too little number

even for optimal game

Next time, please use the question label for a question, the issue label is reserved for problems with the kata itself.

And to answer your question : keep in mind that both players play optimally. For example, with n = 69, Bob can constrain Alice to take only one card each time it is her turn. The beginning of that game is:

• Alice takes 1 card
• Bob takes 1 card
• ....

Random Tests
Small Tests
For n = 69 : expected 30n to equal 8n
Completed in 1ms
Medium Tests
For n = 85853 : expected 42906n to equal 26n
Completed in 1ms
Performance Tests
For n = 428708124794396147 : expected 107177031198598953n to equal 88n
how is it possible?

It tells you right in the second sentence that the number of cards goes up to 10**18 though?

Edit: I've added the performance tag to be even more clear.

Putting a performance test at the end trully is a dick move. At least it should be announced in the instructions.

Impressive, but absolutely unreadable =)

This is a great example of "this works but please never do that".

Damn, forgot my console.log... worst practice guys !

Neither did I. It's nice !

Simple, clear and readable. Nice one !