Noon was my worst hour. Then I counted.
The one pattern in my games that passed the test. Until ten more games came in.
I log every Overwatch match I play. That's 3,680 of them since November 2024. One of the things I write down is the hour.
For months I was sure about my hours. I played best at 9 AM and 2 PM. I believed it the way people believe in lucky socks. Then I had enough games to check. 2 PM turned out to be one of my worst hours: 45.7% over 223 games.
But one hour really did stand out.
At noon I win 39.5% of the time, over 210 games. My overall rate is 48.34%.
That's a nine-point gap. It looks enormous.
Twelve dice
Here's the problem. I didn't test one hour. I tested twelve.
Roll twelve dice. One of them will probably come up low. That doesn't make it a loaded die. If you go looking for the worst of twelve, you will always find one.
The fix is called a Bonferroni correction, and it's simple. If you run twelve tests, make each one twelve times harder to pass. The usual bar is a 5% chance of fooling yourself. Split across twelve hours, it becomes:
0.05 ÷ 12 = 0.0042
A result has to be that unlikely by luck before I call it real.
It passed. Then it didn't.
On September 9, noon stood at 77 wins in 200 games. Depending on exactly how I ran the test, it either just cleared the bar or missed it by a hair. I took the pass. I thought I'd found the one real pattern in my data. I even planned a blog post around it.
Then I kept playing. Ten more noon games. I won six of them.
Noon now sits at a p-value of about 0.0105. For a single test, that's still unlikely. But it's more than twice the bar. It no longer passes.
Nothing about noon changed. Ten games came in, and a borderline result slid back over the line. That's what noise does as the sample grows. The extremes drift back toward the average.
The other end of the chart gets the same treatment. 5 PM is my best hour: 56.0% over 182 games. Its p-value is about 0.04. Not proven either.
What I actually learned
"It passed the test" is a statement about one sample, on one day. It isn't a permanent fact about the world. A number that clears the bar today can fall back under it next week.
So I re-check a finding the day I publish it, not just the day I find it. That includes this post. Every number here was re-run on September 30.
And yes, noon is still my lowest hour. 210 games. I still queue it.