Question regarding SH NL sample sizes

K, I'm not really sure what I'm trying to say, but I'll try anyway.

What do you figure is an appropriate sample to determine whether you were just on a heater, or just running cold etc...

I'm really trying to determine what my leaks are etc. and if I'm just running cold or had just been running hot. i.e. First 20k hands I ran at 7BB (as per Pokertracker) at 100/200 NL. My last 12k hands I've run at -3BB hand but this has been mostly at 200NL and 400 NL.

Can I take any conclusions from this? How many thousand hands would you say are needed to give yourself an idea of how winning/breakeven/losing a player you are shorthanded where the variance is higher, especially if you play a relatively Lag style of play. (I run at 26/16 6-handed). I've heard stuff from 20to 100k hands.

Thoughts?

12k can be a simple bad run or break even play with a few bad decisions.
How would tightening up for a few k hands effect your winrate do you think. I tend to camp a bit more if i feel im running badly to see if it is that or if im just making one or two bad decsions.

Define winrate (w) ptBB/100, standard deviation (sd) ptBB/100.

Your winrate over 100N hands lies within [w-2*sd/sqrt(N), w+2*sd/sqrt(N)] with 95% probability.

For example, if your winrate is 3 ptBB/100 with a sd of 40 ptBB/100 over 12K hands this range is [-4.3;10.3].

20K hands -> [-2.7, 8.7]
40K -> [-1.0, 7.0]
100K -> [0.5, 5.5]

There are some assumptions to this approximation but it gives you an idea about the huge variance within NL Hold'Em.