what is the minimal material odds that it is possible to win with no sacrifices?

[Uri Blass]

I wonder what is the minimal material odds that it is possible to win without sacrifices regardless of what the opponent does.
For the definition of sacrifice suppose we use the values 1,3,3,5,9 for the pieces when I have 2 ideas to define a sacrifice:

1) I can define sacrifice by the following way a move that allow the opponent to win material even after your move (so usually allowing some bad capture when you practically capture back more material in the next ply is not a sacrifice).

2)Second question is the same but sacrifice is only relative to the initial position so if you start without a queen then it is enough for you to be at least a queen up after every move you play and give checkmate to claim you won without sacrifices(it is easier than the first option because in the first option if you have some tactics that win a rook but later lose a knight because of some fork you cannot use it to achieve the target).

I guess it is possible to try to guess the answer by some engine matches when the target of one engine is to win with no sacrifices and the target of the opponent is only to win material.

[Dann Corbit]

Engines see so deeply today that what was a sacrifice before is now purely tactical. Because neural nets do not reveal what their nodes mean, I do not think we can understand if an NN program is doing a speculative sacrifice or if it sees a tactial material grab.
If there is an engine with hand tuned evaluation, then we could examine the reason for moves and know if a move is sacrificial.
On the other hand, is something like a big score for extreme king safety danger or multiple dangerous passed pawns different than wood counting?
I think that if a clear definition of exactly what is a sacrifice and what is not a sacrifice is programmed into the eval as a boolean, it could be added as a value in the search tree and we could track sacrificial wins easily.