Standard deviation number to determine a non random event

[Guest]

We know that standard deviation(sd) is a measure of fluctuation away from the mean.
Good or bad luck go between +/- 3 sd.
When we want to test if a system we use is a winning system we must test samples and count the sd number.
Supose we have data and test +3sd in case we play the 2nd dozen after a previous 1st dozen, and 3rd dozen after the 2nd, you play every spin on the next dozen. It is an hypotetical example.
On a singlezero european wheel we have -2.7%
We have got a 3000 sample. We tested this sample and someone who played 1st dozen after a 2nd dozen had a luck of +3sd
As we found it out after studying the 3000 sample our +3sd we realize that is not the same to pick a dozen at the beginning than scanning which was the best performer. So, this +3sd surely has some added fluctuation(type error 1). My first idea is to substract from 1 to 1.5 sd from the 3 to eliminate regular fluctuation. So, our actual sd number might be 1.5 to 2.
We take our 3000 sample and divide it by 1000 to count their sd in each of the smaller samples. We had +1.8 +1.5 and +2.1sd on each of the 3
The whole data(3000) +3sd on a 12 number play. The same data cut in 3 yielded lower sd on each as expected.

We make a new test , same dozen after dizen to play. We are picking our play beforehand.

We collect the first 1000 trials.

Any player who has played our choice reached +2 sd. The chance to be random seems to be 1/20(95%)

What does it mean?
What is the chance of this last test to be random?
What is the difference between +2sd in 1000 and 2sd in 500 or 5000 trials?
What if next 1000-spin- test yields +2sd?

I apprecciate any answer

ybot