Skip to content

mfgQC

Auditable SPC, capability, and gage R&R for manufacturing — by Brantner Solutions.

PyPI version Python versions License: MIT tests

mfgQC is a quality-control library for manufacturing practitioners — not statisticians or programmers. It does the statistics, data handling, and canonical charts so you can bring the domain knowledge. Every number it reports is checked against its own assumptions and carries a verifiable record of how it was computed.

pip install mfgqc

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.capability())     # Cp/Cpk + Pp/Ppk + an assumption report
print(qc.control_chart())  # the right chart for your subgroup size, with run rules
print(qc.gage_rr())        # ANOVA gage R&R with the AIAG verdict

Why mfgQC

Statistical guardrails

Every analysis checks its own assumptions and reports the result. It warns and recommends — it never silently switches methods. Auto-correction is opt-in.

Practitioner-oriented

No stats or programming background assumed in the public surface. Errors say what's missing and why, naming the exact role or spec limit an analysis still needs.

Auditable by construction

Data and result objects are immutable and carry a hash-chained provenance history, so the full lineage from raw data to final number can be reconstructed and verified.

What it covers

Area Methods
Capability Cp/Cpk (within-σ), Pp/Ppk (overall σ), Cpm, with CIs; Box-Cox / Clements / Johnson for non-normal
Control charts I-MR, X̄-R, X̄-S, p/np/c/u, EWMA, CUSUM, short-run; Nelson & Western Electric run rules
Measurement systems ANOVA gage R&R, bias, linearity, stability, attribute agreement (kappa)
Hypothesis testing assumption-routing t / variance / proportion tests, ANOVA, post-hoc, non-parametrics
Regression & DOE OLS, model selection, logistic, NLS; full/fractional factorials with alias structure
Power & sampling sample size for t / ANOVA / proportion / variance; ANSI/ASQ Z1.4 & Z1.9 acceptance plans
Reliability life-distribution fitting with censoring, Kaplan-Meier, system reliability, MTBF, availability

The differentiator: auditability

Open qcc or Minitab and the lineage from raw data to the reported Cpk lives in your memory and your spreadsheet. In mfgQC it lives in the result object:

cap = qc.transform("boxcox").capability()
[s["operation"] for s in cap.lineage()]
# ['load', 'spec', 'transform', 'capability', 'assumption:normality']

digest = cap.provenance_digest()   # SHA-256 over the computation
cap.verify_provenance(digest)      # True — flips to False if any step was edited

This is what makes the docs you're reading themselves auditable: every method in the Reference states its formula, its assumptions, and the source standard it is pinned to.

  • :material-rocket-launch: Get started — install, load data, run your first analysis.
  • :material-book-open-variant: Reference — the formula, assumptions, and citation behind every method.
  • :material-shield-check: The audit workflow — record, export, and verify a result's lineage.
  • :material-github: Source on GitHub — MIT licensed.