it’s not because of ego—it’s because if checks/simulations are relatively equal in each method, the only way to determine which one is better is by relying on the average number of victories, which is the truly important point here. Since the overall average does not interfere, it balances out proportionally.
Avg checks / wins = X checks / win.
You see X going down, which is great.
You also see wins go up, which is even greater.
However the cost is the same (and I mean identical), because what we have is:
X * win_rate = average cost
But your important points are to have a very small X with a very high win_rate.
So what you did, is to compare a very amplified winner with a very diminished less winner.
Now -
why not apply the same logic for the losses to see who loses more?
Avg checks / losses = X checks / loss.
We're still interested to have a low X even if we lost.
Now the bad news: precise loses more here: it has a much higher X then sequential's X.
So you only see the half of the water glass that you want to look at.
When in fact, if the costs are identical, it has zero relevance who won more, or who lost more, or what the number of ties even was. Those are not metrics, they are betting gambles.
Because, on average, they both do the exact same thing, in the exact same time.
And the best part: it happens even without actually checking for a prefix. The graphs clearly show this:
- when prefix wins: it wins by a very very little difference (differences histogram on the prefix wins are concentraded towards 0)
- when prefix loses: it loses by a very high difference.
Now, here's the conclusion: you are right.