What the tool does
Capability studies for customers and audits come down to a few charts and numbers, but the numbers depend on choices that tools make silently: how sigma is estimated, whether it is unbiased, which rules flag a point. Here each choice is visible.
- Data: paste one subgroup per row, a subgroup column with a value column, or one column with a fixed subgroup size. Missing values (empty, *, N/A) are skipped. Several characteristics in one study.
- Control charts: Xbar-R, Xbar-S and I-MR for measurements, p, np, c and u for counts, with limits for unequal subgroup sizes. Western Electric or Nelson rules mark the points that break them.
- Capability: Cp, Cpk, Cpl, Cpu with within-subgroup sigma, Pp, Ppk and Cpm with the overall standard deviation, confidence limits, expected and observed ppm, histogram with both curves and the specification limits.
- Why numbers differ: the same data with the pooled standard deviation, R̄/d2, s̄/c4 and the spreadsheet STDEV side by side.
- Normality: Anderson-Darling test and normal probability plot.
- Checks and plain sentences: few subgroups, unequal sizes, values outside the specification while Cpk claims capability, limits that look mistyped, and a sentence that says what the numbers mean.
How to use it
- Name the characteristic and enter LSL, USL and the target.
- Choose the layout and paste the measurements.
- Read the control chart first: capability means little if the process is not stable.
- Read the capability tab, compare the estimators if a customer reports another Cpk, and check normality.
- Export the report PDF, the workbook with formulas, the results or the chart.
How it was checked
The control chart constants are computed from the distribution of the range and the standard deviation and match the NIST e-Handbook table of A2, D3 and D4 for subgroups of 2 to 10 to three decimals, with c4(10) = 0.9727. Mean and standard deviation match the certified values of the NIST statistical reference datasets (Michelso, Lew and NumAcc4). The capability example of the NIST e-Handbook 6.1.6 (Cp 1.0, Cpk 0.6667) and its table of rejects (Cp 1.00: 0.27 %, Cp 1.33: 64 ppm) are reproduced, as are the chi-square points used for the Cp confidence limits.
Limits
Gauge R&R, Type 1 studies, acceptance sampling, non-normal capability (Box-Cox, Weibull fits) and AQDEF files are not covered yet. The judgement of capability is the responsibility of the quality engineer and the agreement with the customer.
Last checked 2026-10-20