mfgQC

mfgQC is a quality-control library for Python, written for manufacturing practitioners rather than statisticians. It runs process capability, control charts, gage R&R, and the rest of the SPC toolkit. Every analysis checks its own assumptions and reports them, and every result keeps a verifiable record of how it was computed.

Install

pip install mfgqc

Requires Python 3.10 or newer. It pulls in NumPy, pandas, SciPy, Matplotlib, statsmodels, and scikit-learn.

A first analysis

Load a tidy table, attach the spec limits, and run an analysis. Every result has the same surface: .report() for text, .summary() for a flat dict, .to_dict() for the full payload, and .view() for the chart.

import pandas as pd, mfgqc

qc = (mfgqc.load(df, measure="width", subgroup="lot", subgroup_size=5)
           .spec(lower=1.0, upper=2.0, target=1.5))

print(qc.control_chart())   # the right chart for the subgroup size, with run rules
print(qc.capability())      # Cp, Cpk, Pp, Ppk, plus an assumption report

What it is built on

1. Statistical guardrails. Every analysis checks its own assumptions and reports the outcome. It warns and recommends, and it never silently switches methods. Auto-correction is opt-in.
2. Practitioner oriented. You bring the domain knowledge. mfgQC brings the statistics, the data handling, and the canonical charts. Errors say what is missing and why.
3. Auditable by construction. Data and result objects are immutable and carry a hash-chained provenance history, so the path from raw data to a final number can be reconstructed and verified.

What it covers

Area	Methods
Capability	Cp/Cpk, Pp/Ppk, Cpm with confidence intervals; Box-Cox, Clements, and Johnson for non-normal data
Control charts	I-MR, X-bar R/S, p/np/c/u, EWMA, CUSUM, short-run; Western Electric and Nelson run rules
Measurement systems	ANOVA gage R&R, bias, linearity, stability, attribute agreement
Hypothesis testing	assumption-routing t / variance / proportion tests, ANOVA, post-hoc, non-parametrics
Regression and DOE	OLS, model selection, logistic, non-linear least squares; full and fractional factorials
Power and sampling	sample size for t / ANOVA / proportion; ANSI/ASQ Z1.4 and Z1.9 acceptance plans
Reliability	life-distribution fitting with censoring, Kaplan-Meier, system reliability, MTBF, availability
Bayesian analytics	posterior capability, proportion and rate, and comparison; assurance sample size, guardband, Phase-1 monitoring; hierarchical pooling, censored and short-run

Where to start

• The Quickstart takes you from a table to a result. Every page in the User Guide is a runnable notebook you can open in Colab.
• The Reference gives the formula, the assumptions, and the source standard behind each method.
• The source is on GitHub and the package is on PyPI. MIT licensed.