The computer does not know how to win an endgame on analysis board (multiple lines bug)

In my way to exercise a basic endgame with the practice tool availalbe on the analysis board: lichess.org/analysis/5k2/8/8/8/1P6/8/8/3K4_w_-_-_0_1

I was shocked when the analysis gave me 0 after I played what it is considered to be a winning move (image 1: 7. b5+ Ka7). Interesting enough, I disabled the practice mode so I could see what the computer says about the position.
Apparently, when the computer evaluation is enabled indicating multiple lines (but not 1), it recommends you to play the dead drawn moves (image 2), when Kc7 is clearly winning.

Image 1:

Image 2:

I made this topic because it is interesting to notice the way it works, the giant tablebase lichess provides for endgames is more than enough, there's no need for a local computer evaluation.

Have a nice day!

It's the issue with cloud evaluations.

The "learn from your mistake" feature also gives weird advices on certain endgames, when evaluations shift.
In your case, maybe a threefold repetition after Kc7?

@AACtrl That's an unique position for that setup. Kc7 was not played before. Threefold repetition cannot occur.

Mabye kc7 is not the best move. After 1. kc7, ka8 2. b6 its stalemate. After 1. b6, kb8 2. b7, ka7 3. kc7, ka6 4. b8 = Q white is clearly winning

@dannyo
I appreciate your effort, but after
8. Kc7 Ka8, you go
9. Kb6 Kb8
10. Ka6 Kc7 (if 10. ...Ka8, the black king trying to go into the corner, you just push the pawn all the way up)
11. Ka7 (easy win for white)

The second variation is also not correct from a winning standpoint.
If 8. b6 there's a draw continuation for black: 8. ...Ka8 (instead of your proposed move Kb8).
Whether you move the pawn or 9. Kc7 for the next move, you make a draw.

8. Kc7 is the best move, the computer displays it correctly when choosing to show only 1 line in the evaluation. As mentioned, the tablebase is also an option to check the best moves, since it's all pre-calculated.