What the probability plot shows
Each point is one value plotted against the percentage of values below it, on a scale where the chosen distribution is a straight line. Points that follow the line mean a good fit. A curve means skewness, an S shape means heavy or light tails, a step means rounded data or a gauge with too coarse resolution.
Distribution fitting
The tool fits normal, lognormal, Weibull, gamma and exponential distributions by maximum likelihood and ranks them by the Anderson-Darling statistic A² (smaller is better). The fits are validated against SciPy, see the validation report.
Frequently asked questions
›How do I read the Anderson-Darling p-value?
The test assumes the data are normal. A p-value below 0.05 means that assumption is unlikely, so the data are treated as not normal. A p-value above 0.05 does not prove normality, it only means the data do not contradict it.
›How many bins should a histogram have?
The tool uses about 1.4 times the square root of the number of values, between 6 and 30 bins. With too few bins the shape is hidden, with too many the histogram looks noisy. You can set the number manually.
›What do skewness and kurtosis tell me?
Skewness above 0 means a longer tail to the right, typical for flatness, runout or concentrations near zero. Excess kurtosis above 0 means heavier tails than the normal distribution, below 0 lighter tails, as in a mixture of two processes.
›My data are not normal, what now?
First check whether the process is stable and whether two processes are mixed, for example two cavities or two machines. If the characteristic is naturally skewed, use the distribution fit and the percentile method in the process capability study.