Improving LOS calculation

[voyagerOne]

Instead of looking at stats of individual games...a better idea might be to look at the rounds:

1 Round = 2 games (Same opening, swap colors)

So if A wins 1 game and B wins 1 game. It will be a draw. Score=0

If A wins 1 and draws 1 it will Score 1 point.
If A wins both games it will Score 3 points.

The idea here is if each side won 1 game in the round. It maybe that the opening was superior not the player/engine. If A wins both games...than it should gain a bonus.

[Laskos]

Instead of looking at stats of individual games...a better idea might be to look at the rounds:

1 Round = 2 games (Same opening, swap colors)

So if A wins 1 game and B wins 1 game. It will be a draw. Score=0

If A wins 1 and draws 1 it will Score 1 point.
If A wins both games it will Score 3 points.

The idea here is if each side won 1 game in the round. It maybe that the opening was superior not the player/engine. If A wins both games...than it should gain a bonus.

To understand a bit:

So if A wins 1 game and B wins 1 game. It will be a draw. Score=0 for both A and B.

If A draws 2 games, Score=0 for both A and B.

If A wins 1 and draws 1 it will Score 1 point for A and -1 point for B.
If A wins both games it will Score 3 points for A and -3 points for B.

Correct?

Then say we have 50 Rounds of 2 games each.

15 of them are double wins of A, so a sub-total of +45 points A, -45 points B.
35 of them are B win and a draw, so a sub-total of -35 points A, +35 points B

Total in 50 Rounds: +10 points for A, -10 points for B.
In our usual counting, A has 30 wins and 35 draws, 47.5 points. B has 35 wins and 35 draws, 52.5 points.

So, according to your count, A=10, B=-10, A is stronger.
Normal count, A=47.5, B=52.5, B is stronger.

It's easy to construct outcomes with counting schemes agreeing, say A is stronger in both cases. Therefore this counting scheme is not a monotonic function of the old winning score percentage (translated fairly well via logistic or whatever to some Elo-like rating). Your scheme is not invertible to the old counting (0, 1/2, 1) over entire domain. Might be a good idea, but it has to start from the ground level, nothing like a rating model and scheme is available for it.