More on zeroing the result, which is my recommended approach. This is a theoretical answer.
Assuming an attacker knows how you reset the lock by either zeroing, setting to any fixed value, or scrambling the digits, they should still keep zero knowledge of the correct combination and thus equal odds of matching a random combination.
This could be broken with "mashing around" because no human is a perfect source of random source. Actually they could be the worst.
Mashing around the digits could work with a mechanical/electronic device that spins the digits based on a truly or good-random source.
But normally humans would apply the digits patterns that may reduce the possible values to look for.
Suppose you and the attacker share a set of locks of which both know the combination. Normally one would for example swipe the fingers "randomly" on the reels to make them point to a different number. Or move the reels in an order that the brain wants to keep.
Maybe somebody will make sure the resulting number shows all digits different from the correct combination, or a minimum number of ticks when changing each digit.
This will result in a known plaintext attack of an increasing number of attempts (again, this is a theoretical answer) and will give additional information on the combination that the attacker should not have.
What does emphasized additional mean? That even if the attacker succeeds in determining that a single digit is surely a wrong guess they have just dropped the needed brute force attacks by 1000. Add more digits to restrict the attack surface.
Setting to 0000
or to any predefined value makes the odds of every combination the same