Below half?
Last week I couldn't prove I was worse than a coin. 41 games later, I can.
I've played 3,680 matches of Overwatch since November 2024. I wrote down every one.
It's supposed to be close to half
Overwatch's matchmaking tries to give you even games. Win a lot, and it finds you tougher opponents. Lose a lot, and it eases off. A system like that pushes everyone toward 50%.
So a number near half isn't a verdict on me. It's the machine doing its job.
Is my win rate even below half?
It feels below. An even record would be 1,840 wins. I'm 61 short.
But run the test. Suppose I really were a coin flip. How often would 3,680 games land this far from half, just by luck?
About 4% of the time.
The usual bar for "this isn't luck" is 5%. A week ago, at 3,639 games, it was 6% and I missed. 41 games later it's 4.4%. I clear it. Barely.
So why log anything?
If the win rate is pinned near half by design, the win rate was never the point. The point was the stories I told myself about it.
Gibraltar was "my map," at 68%. Over 147 games it's 55.1%, and across 32 maps no difference survives. I "played best" at 2 PM. It's below my average. Noon looked cursed, until ten more games came in.
Every one of those felt true. The record is how I found out they weren't.
A weak learner, technically
A weak learner is a rule that's right a bit more than half the time. At 48.34%, I'm not even that.
But boosting has a rule for this. When a learner is wrong more than half the time, AdaBoost doesn't throw it out. It flips its vote. A rule that's wrong 51.66% of the time becomes, flipped, a rule that's right 51.66% of the time.
So if you ever need a weak learner, bet against me. You'd win 51.66% of the time.
Just don't bet much. Last week this same test said 6%.