Each column is independent, so we can analyze first just one of them.
After a few cases, with any initial digit guaranty to win in 4 tries, but the better first number are rot13(barguerrfrirabeavar) because average of the expected number of attempts is 2.6 but with the other digits it's 2.7.
The difference is bigger when you consider the 5 columns, because the slowest column ruins the game. With the good initial digits you have a 41% chance of using 4 attempts and with the bad 67%. The average of the expected numbers of attempts is 3.4 vs 3.7.
(If you program a version with more columns, he difference increase even more.)
My recommendation is to start with rot13(nyybarfbenyyavarf) because it makes analyzing the second step much easier, but any random combination of the good digits is a good first number.
I got that 59% of the games are solvable in 3 or fewer guesses using the optimal strategy. (I'll take a look at my calculations later.) All my cases have some tricky part, so I had to use a piece of paper to analyze what to do in the next step.
I found also a simple strategy that involves playing always fixed numbers in the fist and second step, and then thinking a little for the third step. This is easier but guaranty only a 32% win rate in 3 or fewer guesses, and 100% in 4 or fewer guesses. (But I didn't follow alllll the cases for this strategy, there is a tiny chance that I missed a better combination of initial numbers.)
As I said, it's an interesting game and I had to use some time to get the optimal strategy. It's not too easy, but not too hard.
PS: I second the feature request of rawling to move the 0 to the right.
PS2: It would be nice that "close" and "warm" has more contrast.
Wordle, but for numbers. Guess the 5-digit code in 6 tries. Each guess reveals how close you are - but not which direction. New puzzle daily. Can you crack today's Numble?
[spoiler alert]
Each column is independent, so we can analyze first just one of them.
After a few cases, with any initial digit guaranty to win in 4 tries, but the better first number are rot13(barguerrfrirabeavar) because average of the expected number of attempts is 2.6 but with the other digits it's 2.7.
The difference is bigger when you consider the 5 columns, because the slowest column ruins the game. With the good initial digits you have a 41% chance of using 4 attempts and with the bad 67%. The average of the expected numbers of attempts is 3.4 vs 3.7.
(If you program a version with more columns, he difference increase even more.)
My recommendation is to start with rot13(nyybarfbenyyavarf) because it makes analyzing the second step much easier, but any random combination of the good digits is a good first number.