If the picture is legible it should be fairly self explanatory.
I wonder why the Bell curve is so one sided.
Is a sample of thirty seven enough?
Is a half grain going to make any difference at short range? It is 1% of 50 grains.

http://i88.photobucket.com/albums/k191/KGKILBY/SAMPLE.jpg

37 is not a statistically significant sample. 1/5 gr. is not going to affect shooting to any noticeable degree. Richard Corbin (RCE, Inc.) considers 1% as a perfectly acceptable range for swaged bullets. When casting, just a few degrees difference in mould/alloy temperature will make more difference than this, even when bullets are otherwise perfect.

If you want to apply statistical methods, you have to start with the rate of heating of your furnace. You'll find that the variation here may be significant. If you have the equipment for measuring, you'll find that the rate of heat-up will vary markedly. Nevertheless, rather than approach this statistically, just select those bullets that vary less than 1/5 grain (ALK8944).
johnjay

A sample size of thirty is an old statistician's rule of thumb for a minimum number that is likely to give you a symmetrical bell curve sampling a random source of data. Since you didn't get a bell curve in your histogram, it is likely the weight data isn't random. That can be a good clue for process refinement.

I remember a textbook case taken from real life where a printer who was paying for high grade paper kept having his equipment foul up, due mostly to paper thickness adjustment problems. He hired a statistician who found the weight of randomly selected samples would climb up like a bell curve, then abruptly crash straight down, like falling off a cliff, have nothing until he got to the other side of the bell curve, where it formed a mirror image of the first half. In other words, it made a bell curve whose middle section was cut out missing. The cause? The paper supplier was selecting the best weights out for a preferred customer and letting this printer pay full price for the tail ends of the curve.

In your case, I would suggest you are getting those heavier bullets because the mold hasn't closed quite 100%. Usually little bits of flashing or stray dross dust cause that. That's a one-way problem, as the mold halves can't over-close and make the cavities too small. Hence, the asymmetry of your histogram curve, as that failure to close completely introduces a condition not present for most of your castings. If you can get the blocks closed perfectly every time, you will get pretty close to bell curve symmetry.

The 30 sample number is a minimum for likely symmetry, but a better rule for large quantities, to establish confidence in what you produce, is a sample equal to the square root of the number you intend to produce plus 1, and round up. So, for 1000 bullets, that is 33 randomly selected samples. For 10,000, it is 101. I saw a military document once for producing parts that were not to be individually tested because testing destroyed them. If 86 randomly selected samples all worked perfectly, they considered that they had a high enough confidence level to accept the whole lot.

Since this old thread got bumped to the top I will show a graph I did subsequent to the one in the original post. It seems to show a variance between each of the two cavities.

I usually throw out obvious defects as I cast; sometimes one cavity or the other seems to get on a "bad streak" where it produces visually defective bullets. I wonder if in the first graph I didn't have an inordinate number of bullets from only one cavity.

In this application sorting the bullets by weight made no measurable difference in group size.

http://i88.photobucket.com/albums/k191/KGKILBY/Bell002.jpg