A guided workflow, with visible assumptions
This first workbench supports continuous measurements in one column, or equal-size rational subgroups of 2–10 measurements in separate columns. Confirm the separator, decimal mark and header before choosing measurement columns. Missing, invalid or non-finite selected cells stop the analysis with their row and column; they are never silently removed.
A report tied to one confirmed dataset
The method passport records this local dataset revision, chosen columns, CSV record references, unit, collection-order and subgroup confirmations, exact estimator constants, Phase I baseline and the single signal rule. Edits invalidate the prior report. Reinspect after changing imported data, separator, header or decimal settings. Record numbers are parsed CSV rows, not physical lines when a quoted cell contains a line break. This local revision is not persistent storage, regulatory traceability or an audit certificate.
Why Cp/Cpk can differ between software
Here, individuals use the mean moving range of two consecutive readings divided by 1.128. Subgroups use the mean range divided by d₂ for their size. Cp/Cpk use this within-process estimate; Pp/Ppk use the sample standard deviation across all measurements, with divisor N−1 and no c₄ correction. Different grouping, order, sigma estimators, bias corrections and rounding can change results. Match these settings before comparing software.
Control limits are not specifications
Control limits describe estimated process behaviour. Specifications describe requirements. A point may meet its specification while signaling a process change, or miss a specification without triggering a control-chart signal. This workbench checks only points outside three-sigma limits; absence of such signals does not prove stability. All rows form a preliminary Phase I baseline.
Read the result before making a decision
Capability indices are withheld unless you explicitly select the normal-model assumption. When displayed, they are descriptive estimates, not a certificate that production is capable. Investigate special causes, dependence, distribution shape and the measurement system first. No predicted defect rate or automatic pass/fail decision is given. One-sided specifications produce only the available one-sided indices; no two-sided Cp/Pp is invented.
Use the existing I-MR tool or X̄-R / X̄-S tool for more signal rules and baseline options, and the capability calculator for additional estimators. Data are not automatically transferred to these pages.
Zero observed variation
If all readings are identical, or every subgroup is internally constant, the within sigma estimate is zero. The workbench still reports observations, sample summaries and actual specification counts. It does not divide by zero or present collapsed control limits as evidence of stability: plots show observations and their center only, signal tests and the capability comparison are withheld. When subgroups differ, overall sample SD may remain positive; withholding this comparison is a conservative workflow restriction, not a claim that every overall index is mathematically undefined. Review measurement resolution and subgroup design. Nonidentical values whose variation underflows numeric precision still stop analysis rather than being treated as observed constancy.
Sources and scope
Methods follow the NIST Shewhart control-chart guidance and NIST process capability guidance. This is a focused workflow, not a replacement for all Minitab analyses. No server upload, account or cloud storage is required; closing or reloading clears the workbench data.
Frequently asked questions
Is my CSV uploaded?
No. This workbench reads the selected file inside your browser and does not save it in a URL or local storage.
Are limits and specifications the same?
No. Control limits are estimated from measurements; specifications are engineering or customer requirements.
Why is capability hidden?
Confirm the normal-model assumption and provide at least one specification limit. Invalid data prevents analysis. With zero within variation, observed measurements, sample summaries and specification counts remain available, but this workflow withholds control limits, control signals and the capability comparison. Zero observed variation does not prove stability or zero future risk.
Can I use unequal subgroups?
Not in this first workflow. Use the separate capability and control-chart tools, checking their method and assumptions.
Does no signal mean the process is stable?
No. Only the point-beyond-three-sigma rule is checked here. Additional patterns, dependence and process knowledge need review.