Once you establish a mean value, there is still natural variation around the mean. That's the 2-3 hp you're talking about, but "natural variation" means that one run it could be 2-3 hp below mean, and 2-3 hp above the mean the next time.
Just because a data point falls within the upper control limit (typically 3xSTD DEV) doesn't mean it's "natural." If the data starts falling in the upper control range, and stays there, then you have to calculate a new baseline average (mean).
