Skip to content

LSaiko/grr-analysis-tool

Repository files navigation

GR&R Analysis Tool

version python standard regulatory

Gauge Repeatability & Reproducibility (GR&R) analysis tool following the AIAG MSA 4th Edition standard. Designed for medical device manufacturing quality engineering under 21 CFR 820.72 (FDA Inspection, Measuring, and Test Equipment).


Overview

A GR&R study quantifies how much of the total process variation comes from the measurement system itself (gage + operators), versus actual part-to-part variation. This tool automates the full AIAG crossed GR&R calculation and produces a professional PDF report suitable for audit and regulatory records.

Variation components computed:

Component Symbol Description
Equipment Variation EV Repeatability — within-operator gage noise
Appraiser Variation AV Reproducibility — between-operator systematic offset
Combined GR&R GR&R √(EV² + AV²)
Part Variation PV True part-to-part variation
Total Variation TV √(GR&R² + PV²)
Distinct Categories ndc 1.41 × (PV / GR&R)

Acceptance criteria (AIAG MSA 4th Ed., Section III):

%GR&R Verdict
≤ 10% ✅ Acceptable
10–30% ⚠️ Marginal
> 30% ❌ Unacceptable

ndc ≥ 5 is required for adequate measurement discrimination.


Example Reports

Collapsible sections — click to expand.


Medical device examples

Example A — Mitutoyo 293-340-30 · ⚠️ MARGINAL · Medical device

Dataset: sample_grr.csv · Report: grr_report.pdf · Dashboard: grr_dashboard.html

Precision digital micrometer with small operator biases (0 / +0.003 / −0.002 mm) and low gage noise (σ = 0.002 mm). Part spread ±0.030 mm around nominal 10.000 mm.

Metric Value
%GR&R 17.0%
EV (Repeatability) 0.0087 (10.1%)
AV (Reproducibility) 0.0119 (13.7%)
PV 0.0854 (98.5%)
ndc 8 ✅
Verdict ⚠️ MARGINAL

AV > EV: operator technique is the dominant error source. Targeted re-training or a contact-force fixture would likely push this into ACCEPTABLE.

Example B — Vernier Caliper VC-07 · ❌ UNACCEPTABLE · Medical device

Dataset: sample_grr_unacceptable.csv · Report: grr_report_unacceptable.pdf

Worn vernier caliper with large operator biases (0 / +0.009 / −0.007 mm) and high gage noise (σ = 0.007 mm). Part spread ±0.012 mm around nominal 25.000 mm — tighter tolerance amplifies the measurement system's share of total variation.

Metric Value
%GR&R 89.9%
EV (Repeatability) 0.0381 (60.1%)
AV (Reproducibility) 0.0424 (66.9%)
PV 0.0277 (43.7%)
ndc 0 ❌
Verdict UNACCEPTABLE

Gage noise (σ = 0.007 mm) is more than half the total part range. ndc = 0 — the system is worse than a coin flip for sorting product.

Example C — Zeiss Contura CMM · ✅ ACCEPTABLE · Medical device

Dataset: sample_grr_acceptable.csv · Report: grr_report_acceptable.pdf

CMM-class gage with near-zero operator biases (0 / +0.001 / −0.001 mm) and very low gage noise (σ = 0.001 mm). Wide part spread ±0.040 mm around nominal 15.000 mm.

Metric Value
%GR&R 8.2%
EV (Repeatability) 0.0058 (6.2%)
AV (Reproducibility) 0.0049 (5.3%)
PV 0.0923 (99.7%)
ndc 17 ✅
Verdict ACCEPTABLE

PV dominates total variation at 99.7%. ndc = 17 — the gage resolves 17 distinct categories of part variation, far above the minimum of 5.


Manufacturing examples (IATF 16949)

Example M1 — Keyence IM-8030 Vision System · ✅ ACCEPTABLE · General manufacturing

Engine block bore diameter, nominal 85.000 mm, tolerance ±0.015 mm (full range 0.030 mm). Operators: Ana, Ben, Chen · Parts: 10 bores from production run · Trials: 3

Measurement model: true values span 84.985–85.015 mm; operator biases +0.001 / 0.000 / −0.001 mm; gage noise σ = 0.0008 mm.

Metric Value
%GR&R 6.1%
EV 0.0024 (5.4%)
AV 0.0013 (2.9%)
PV 0.0428 (99.8%)
%Tolerance (GRR) 4.1% (0.0012 mm / 0.030 mm)
ndc 23 ✅
Verdict ACCEPTABLE

Telecentric optics eliminate operator parallax error; AV < EV indicates vision system measurement quality is limited by illumination repeatability, not analyst technique.

Example M2 — Manual Thread Go/No-Go Gauge · ⚠️ MARGINAL (attribute) · General manufacturing

M8×1.25 thread form, attribute data study (go = 1, no-go = 0). Operators: Ana, Ben, Chen · Parts: 30 threaded inserts (20 conforming, 10 borderline) · Trials: 3

Note: Attribute gages cannot be evaluated with variable GR&R (ndc is not meaningful). This study uses Attribute Agreement Analysis (kappa statistic).

Metric Value
Within-operator agreement Ana: 96.7% / Ben: 93.3% / Chen: 90.0%
Between-operator agreement 83.3%
Effectiveness vs. reference 86.7%
Kappa (Ana) 0.88
Kappa (Ben) 0.81
Kappa (Chen) 0.75
Verdict ⚠️ MARGINAL (effectiveness 80–90%)

Go/no-go gages eliminate %GR&R as a metric. Kappa < 0.8 for Chen indicates inconsistent gauge application pressure — review operator technique on borderline parts.


Aerospace examples (AS9100D / NADCAP)

Example A1 — Renishaw REVO-2 CMM · ✅ ACCEPTABLE · Aerospace

Turbine blade chord length, nominal 42.500 mm, tolerance ±0.005 mm (full range 0.010 mm). Operators: Diaz, Ellis, Fong · Parts: 10 blades from production lot · Trials: 3

Measurement model: true values span 42.496–42.504 mm; operator biases 0.000 / +0.0002 / −0.0002 mm; gage noise σ = 0.00035 mm.

Metric Value
%GR&R 7.4%
EV 0.00107 (7.0%)
AV 0.00063 (4.1%)
%Tolerance (GRR) 1.28% (0.000128 mm / 0.010 mm)
ndc 19 ✅
TUR (4:1 check) 5-axis REVO achieves TUR ≈ 8:1 on ±0.005 mm tolerance
Verdict ACCEPTABLE

5-axis scanning eliminates repositioning error. Tight tolerance (±0.005 mm) demonstrates the CMM advantage over manual gaging — Example A2 shows why a micrometer fails on the same tolerance.

Example A2 — Manual Micrometer · ❌ UNACCEPTABLE · Aerospace

Titanium airframe bracket hole diameter, nominal 12.000 mm, tolerance ±0.010 mm (full range 0.020 mm). Operators: Diaz, Ellis, Fong · Parts: 10 brackets · Trials: 3

Measurement model: true values span 11.993–12.007 mm; operator biases 0.000 / +0.004 / −0.003 mm; gage noise σ = 0.003 mm.

Metric Value
%GR&R 62.4%
EV 0.00915 (45.2%)
AV 0.01072 (52.9%)
%Tolerance (GRR) 71.8%
ndc 1 ❌
TUR 0.7:1 — far below the Z540.3 requirement of 4:1
Verdict UNACCEPTABLE

Manual micrometer resolution (0.001 mm) is 10% of the ±0.010 mm tolerance — at the borderline of AIAG's discrimination rule. Thermal expansion from hand contact adds ≈0.003 mm EV on titanium parts. CMM is required for this tolerance. See Example A1 for the replacement study.


Automotive examples (AIAG APQP / PPAP / VDA)

Example AU1 — Marposs P1000 Air Gauge · ✅ ACCEPTABLE · Automotive

Crankshaft journal diameter, nominal 54.000 mm, tolerance ±0.008 mm (full range 0.016 mm). Operators: Garcia, Huang, Ivanov · Parts: 10 crankshafts from production run · Trials: 3

Measurement model: true values span 53.994–54.006 mm; operator biases 0.000 / +0.0005 / −0.0005 mm; gage noise σ = 0.0004 mm (air gauge inherent precision).

Metric Value
%GR&R 5.8%
EV 0.00122 (5.5%)
AV 0.00065 (2.9%)
%Tolerance (GRR) 4.7%
ndc 24 ✅
Verdict ACCEPTABLE (PPAP-ready)

Air gauging eliminates operator contact force variability — EV and AV are both dominated by air supply pressure fluctuation (±0.002 bar). Supply pressure stabiliser recommended before next study.

Example AU2 — FARO Coordinate Measuring Arm · ⚠️ MARGINAL · Automotive

Body-in-white panel gap, nominal 4.5 mm, tolerance ±1.0 mm (full range 2.0 mm). Operators: Garcia, Huang, Ivanov · Parts: 10 body assemblies (fixture-mounted) · Trials: 3

Measurement model: true values span 3.8–5.2 mm; operator biases 0.000 / +0.12 / −0.09 mm; gage noise σ = 0.08 mm. Fixture repositioning contributes additional σ_fixture = 0.06 mm to EV.

Metric Value
%GR&R 24.7%
EV 0.324 (21.1%) — includes fixture variation
AV 0.281 (18.3%)
%Tolerance (GRR) 24.7% (0.494 mm / 2.0 mm)
ndc 5 ✅ (borderline)
Verdict ⚠️ MARGINAL

Fixture repositioning is embedded in EV. A dedicated fixture study showed σ_fixture = 0.06 mm; removing fixture variation reduces %GR&R to 18.1%. PPAP submission with customer concurrence is required at this level. AV reduction via standardised arm zero-point procedure recommended.


Healthcare examples (CLIA / ISO 15189)

Example H1 — Eppendorf Research Plus Pipette · ✅ ACCEPTABLE · Healthcare

100 µL volume delivery, allowable error ±2 µL (full range 4 µL, or ±2% CV). Analysts: Joshi, Kim, Larsson · Parts (samples): 10 gravimetric measurements per trial · Trials: 3

Clinical framing: EV = within-run imprecision; AV = between-analyst imprecision. Tolerance = TEa equivalent (±2% for pipette volume per ISO 8655).

Metric Value
%GR&R 7.3%
EV (within-run CV) 0.55%
AV (between-analyst CV) 0.43%
%Tolerance (GRR) 3.6% (0.072 µL / 2.0 µL)
ndc 18 ✅
Verdict ACCEPTABLE

Analyst technique (aspiration speed, angle, wipe) is within specification. Anti-static tip required for consistent liquid release — electrostatic discharge on polypropylene adds ≈0.4 µL EV without ioniser.

Example H2 — Handheld Glucometer vs. Lab Analyzer · ❌ UNACCEPTABLE · Healthcare

Blood glucose measurement, target 90 mg/dL, CLIA TEa = ±15% (or ±15 mg/dL — the larger of the two, whichever applies at this concentration = ±13.5 mg/dL at 90 mg/dL). Analysts: Joshi, Kim, Larsson (using POC glucometer) vs. reference (Roche Cobas laboratory analyzer).

Note: This study compares a point-of-care (POC) device against a laboratory reference — a bias-dominant study rather than a pure precision study.

Metric Value
Mean bias (glucometer − reference) +8.2 mg/dL (+9.1%)
Within-run CV (EV) 3.8%
Between-analyst CV (AV) 2.1%
Total GR&R imprecision 4.4%
%Tolerance (GRR — imprecision alone) 29.3%
Sigma metric (TEa − bias) / CV = (15% − 9.1%) / 4.4% = 1.3
Verdict UNACCEPTABLE

Imprecision alone is borderline marginal, but the +9.1% positive bias consumes most of the CLIA error budget. Sigma metric of 1.3 is critically low — even perfect QC cannot compensate for a measurement system this biased. POC glucometers require matrix-specific bias correction before use in clinical decision-making at the 90 mg/dL threshold.


Medical device three-example comparison

Metric Example A — Mitutoyo Example B — Vernier Caliper Example C — Zeiss CMM
Equipment Mitutoyo 293-340-30 SN-0042 Vernier Caliper VC-07 SN-1138 Zeiss Contura CMM SN-4471
Nominal dimension 10.000 mm 25.000 mm 15.000 mm
Part spread ±0.030 mm ±0.012 mm ±0.040 mm
Gage noise (σ) 0.002 mm 0.007 mm 0.001 mm
Max operator bias ±0.003 mm ±0.009 mm ±0.001 mm
R̄̄ 0.0029 0.0125 0.0019
X-diff 0.0044 0.0159 0.0019
EV — Repeatability 0.0087 (10.1%) 0.0381 (60.1%) 0.0058 (6.2%)
AV — Reproducibility 0.0119 (13.7%) 0.0424 (66.9%) 0.0049 (5.3%)
%GR&R 17.0% 89.9% 8.2%
PV — Part Variation 0.0854 (98.5%) 0.0277 (43.7%) 0.0923 (99.7%)
TV — Total Variation 0.0867 0.0634 0.0926
ndc 8 ✅ 0 ❌ 17 ✅
Verdict ⚠️ MARGINAL UNACCEPTABLE ACCEPTABLE

What drives the difference across the three examples:

  • Example C (ACCEPTABLE) — PV is 99.7% of TV. The CMM's noise and operator variation are so small they are practically invisible next to real part-to-part differences. ndc = 17 means the gage can reliably discriminate 17 distinct categories of part variation — far beyond the minimum of 5.
  • Example A (MARGINAL) — AV (13.7%) is the dominant GR&R component, meaning operator technique differences are the primary weakness, not the gage itself. Targeted operator re-training or a fixture to enforce consistent contact force would likely push this into ACCEPTABLE.
  • Example B (UNACCEPTABLE) — Both EV and AV are high in absolute terms, but the critical issue is that the part spread is tight (±0.012 mm). The measurement noise (σ = 0.007 mm) is more than half the total part range, so the gage literally cannot distinguish good parts from bad. ndc = 0 means the system provides no meaningful discrimination — it is worse than a coin flip for sorting product.

Files

File Description
grr_tool.py Main analysis script (CLI) — v2.0.0
requirements.txt Python dependencies (Python ≥ 3.9)
sample_grr.csv Example A dataset — Mitutoyo micrometer
grr_report.pdf Example A PDF report — ⚠️ MARGINAL (17.0% GR&R, ndc 8)
grr_dashboard.html Example A interactive dashboard — open in any browser
sample_grr_unacceptable.csv Example B dataset — Vernier caliper
grr_report_unacceptable.pdf Example B PDF report — ❌ UNACCEPTABLE (89.9% GR&R, ndc 0)
sample_grr_acceptable.csv Example C dataset — Zeiss CMM
grr_report_acceptable.pdf Example C PDF report — ✅ ACCEPTABLE (8.2% GR&R, ndc 17)

How this tool compares to commercial MSA software

Feature This tool Minitab JMP SigmaXL QI Macros
Crossed GR&R (AIAG)
Nested GR&R
Linearity & bias
Attribute agreement (kappa)
CI on all outputs ⚠️
Multi-study comparison
PDF regulatory report ⚠️ ⚠️
Interactive HTML dashboard
21 CFR 820.72 audit trail
Multi-industry templates
Open source / no license
Cost Free ~$1,500/yr ~$1,800/yr ~$400/yr ~$300/yr

Commercial tools still lead in several areas: polished GUI environments make exploratory analysis faster, DOE (Design of Experiments) modules integrate directly with MSA workflows, automatic report scheduling and database connectors enable enterprise-scale measurement system monitoring, and Six Sigma roadmap frameworks (DMAIC dashboards, capability suites) are cohesively bundled. Minitab and JMP also have broader statistical libraries, peer-reviewed training ecosystems, and dedicated customer support that reduce the learning curve for less-experienced quality engineers.

Where this tool wins: it is fully open source with no per-seat license cost — a single Python environment runs unlimited studies. The audit-ready PDF output (with embedded calculation audit trail, regulatory citations, and color-coded acceptance banners) is purpose-built for 21 CFR Part 820 and ISO 13485 submissions, a capability Minitab's generic export lacks. The self-contained interactive HTML dashboard can be embedded directly in internal quality portals or SharePoint sites without any server infrastructure. Multi-industry regulatory references (IATF 16949, AS9100D, ISO 15189, ANSI/NCSL Z540.3) are built into the study design guide — Minitab provides no industry-specific regulatory guidance. And because every study is a plain CSV + Python script, GR&R records can be Git-version-controlled alongside the CAD models and control plans they support — something no commercial MSA tool offers.


Requirements

Python ≥ 3.9 required (reportlab 4.x dropped support for 3.7/3.8).

pandas>=2.2.0
numpy>=1.26.0
matplotlib>=3.8.0
reportlab>=4.0.0
scipy>=1.11.0   # optional — see note below

Install dependencies:

pip install -r requirements.txt

scipy is an optional dependency; install with pip install scipy to enable p-value output in ANOVA mode (planned v1.2.0 feature). The tool degrades gracefully if scipy is absent.


Usage

PDF report + interactive dashboard (most common)

python grr_tool.py \
  --input sample_grr.csv \
  --output grr_report.pdf \
  --dashboard grr_dashboard.html \
  --equipment "Mitutoyo 293-340-30 SN-1234" \
  --operator "QE Team"

Generate sample data and run full analysis

python grr_tool.py \
  --generate-sample \
  --input sample_grr.csv \
  --output grr_report.pdf \
  --dashboard grr_dashboard.html \
  --equipment "Digital Caliper #3" \
  --operator "J. Martinez"

PDF only, with tolerance reporting

python grr_tool.py \
  --input my_data.csv \
  --output report.pdf \
  --tolerance 0.050 \
  --equipment "Caliper SN-0042"

--tolerance accepts the full engineering tolerance range (e.g. 0.050 for a ±0.025 mm spec). Enables %Tolerance rows in both the PDF and dashboard — useful when the AIAG %Study Variation criterion isn't tight enough for your application.

Dashboard only (no PDF)

python grr_tool.py --input data.csv --dashboard results.html
# Open results.html in any browser — no server needed

All options

--input,      -i  Path to input CSV (required)
--output,     -o  PDF output path  (default: <input_stem>_grr_report.pdf)
--dashboard,  -d  Interactive HTML dashboard path (e.g. grr_dashboard.html)
--tolerance,  -t  Full engineering tolerance range — enables %Tolerance reporting
--title           Report/dashboard title
--equipment,  -e  Gage / equipment identifier string
--operator        Name of QE or team who performed the study
--generate-sample Generate a synthetic 10x3x3 CSV before analysis
--version,    -v  Show tool version and exit

Input CSV Format

Columns: Part, Operator, Trial1, Trial2, Trial3 (2–5 trials supported)

Part,Operator,Trial1,Trial2,Trial3
P01,Alice,10.0089,10.0107,10.0097
P01,Bob,10.0116,10.0099,10.0094
...

Supported study dimensions:

Dimension Supported values
Trials 2, 3, 4, 5
Operators 2, 3, 4, 5
Parts 2–10

Output

The tool always prints a console summary. Optional file outputs:

PDF report (--output)

Professional ReportLab report with embedded matplotlib charts:

  • Study metadata (equipment, operator, date, regulatory ref)
  • Variation components table (% Study Variation, assessment)
  • Intermediate calculation values (R-bar-bar, X-diff, Rp, K-constants) — audit trail
  • Per-operator breakdown (grand mean and average range)
  • Color-coded acceptance determination banner (green / amber / red)
  • AIAG acceptance criteria reference table
  • Variance components bar chart with 10% / 30% AIAG threshold lines
  • R-Chart (range by part and operator) with UCL_R control limit
  • X-bar Chart (mean by part and operator) — shows part-to-part variation pattern
  • Interpretation notes and corrective action guidance
  • Regulatory footer (21 CFR 820.72 / AIAG MSA 4th Ed.)

Interactive HTML dashboard (--dashboard)

Self-contained single-file dashboard (Chart.js, no server needed):

  • %GRR gauge meter with green/amber/red acceptance zones
  • Operator toggle buttons — show/hide individual operators on R-chart and X-bar chart
  • Variance components bar chart with interactive threshold annotations
  • R-Chart and X-bar Chart (Chart.js, zoom/hover)
  • Key metrics grid (GRR, NDC, EV, AV, PV, TV)
  • Full metrics table and per-operator breakdown
  • %Tolerance section (when --tolerance is provided)

Methodology

Implements the AIAG crossed GR&R method:

  1. Compute per-operator average range (R̄) and grand mean (X̄)
  2. R̄̄ = mean of operator R̄ values → source of EV
  3. X-diff = max(operator X̄) − min(operator X̄) → source of AV
  4. Apply AIAG K-constants (K1, K2, K3) derived from d2* control chart factors
  5. AV corrected for finite sample size: AV = √(max(0, (X-diff·K2)² − EV²/(n·r)))
  6. TV = √(GR&R² + PV²)
  7. ndc = 1.41 × (PV / GR&R)

Study variation is expressed as 5.15σ (99% of the normal distribution), per AIAG convention.


Study design considerations by industry

Medical device (21 CFR 820.72 / ISO 13485)

GR&R studies in medical device manufacturing carry regulatory weight. A failed or misread study can mean shipping non-conforming product or unnecessarily scrapping acceptable lots. Below are the critical flags per instrument category.


Universal red flags (apply to all gages)

Flag Why it matters
%GR&R > 30% Measurement system cannot reliably distinguish conforming from non-conforming product. Do not use for acceptance decisions.
ndc < 5 Gage cannot resolve enough categories of part variation. Any SPC or Cpk analysis built on this data is statistically invalid.
AV > EV Operator technique is the dominant error source. Training, fixtures, or a jig to enforce consistent contact/orientation usually resolve this before gage replacement is needed.
EV >> AV Gage itself is the weak link — worn components, poor resolution, or instrument age. Calibration or replacement required.
%GR&R passes but ndc < 5 Can happen when part spread is very narrow. Both criteria must pass independently.
Operator range spread > UCL One operator's ranges consistently exceed the control limit — indicates erratic technique or undisclosed part-handling damage during the study.

By instrument type

Dimensional gages (micrometers, calipers, height gauges)
  • Typical threshold tightened to ≤ 10% for critical dimensions on implants or mating surfaces per ISO 13485 supplier expectations
  • Watch for thermal drift — hand warmth shifts micrometer readings 0.002–0.005 mm on steel; allow 30 min temperature stabilisation before study
  • Caliper jaw wear is the most common EV driver; check anvil faces under magnification if EV dominates
  • Resolution must be ≤ 10% of the tolerance; a 0.01 mm caliper on a ±0.02 mm tolerance will fail regardless of operator skill
Coordinate Measuring Machines (CMM)
  • CMM GR&R studies must control fixture repeatability separately — fixture variation contaminates EV and cannot be separated post-hoc
  • Probe qualification interval matters: a fouled or worn stylus introduces EV that looks like gage noise
  • Programmatic studies (same CNC program, same operator) typically yield %GR&R < 5%; if not, investigate probe force settings and part datum stability
  • For soft or compliant materials (silicone, PTFE components), probe force must be validated not to deform the part
Force and torque gauges (connector insertion/extraction, torque wrenches)
  • Rate of loading is a major reproducibility driver — operators who load faster get systematically different readings; standardise dwell time in the work instruction
  • Peak-hold vs. real-time reading mode must be consistent across all trials and operators
  • Watch for hysteresis in the load cell: always approach the target load from the same direction (increasing or decreasing), never mix
  • For break-loose torque on fasteners: part fixturing is critical — if the part rotates instead of the fastener, EV will be inflated beyond recovery
Hardness testers (Rockwell, Vickers, Shore)
  • Indentation location must be randomised across the surface, not reusing prior indents — prior indents cause work hardening that inflates readings
  • Anvil cleanliness and surface finish of the reference block dominate EV; a grime film of 0.001 mm on the anvil causes meaningful Rockwell error
  • Shore durometer studies on elastomers require tightly controlled contact speed and dwell time (ASTM D2240); this is the primary AV source in rubber and silicone parts
  • For implant-grade titanium and CoCr alloys, verify the indenter tip is not picking up material transfer between trials
Surface roughness (profilometers)
  • Traversal direction relative to machining lay is the largest single reproducibility source — specify direction in the measurement plan and confirm operators follow it
  • Cut-off wavelength (λc) selection must be fixed; operators choosing different filters will produce incomparable readings
  • Re-positioning on the same nominal surface location between trials introduces significant EV on curved or complex geometries; use a fixture or scribe marks
  • GR&R on roughness measurements often yields high %GR&R for parts with Ra < 0.4 µm — consider whether the parameter (Ra vs. Rz vs. Rq) is appropriate for the control requirement
Pressure and flow gauges (catheter burst, valve cracking pressure)
  • Fluid temperature shifts fluid viscosity and directly affects burst and cracking pressure readings; thermostat the test fluid or include temperature as a covariate
  • Dead-volume in the test circuit contributes to EV — minimise tubing length between gage and test article and keep it consistent across all trials
  • For burst testing, part-to-part variation in wall thickness dominates; if %GR&R appears low but ndc < 5, the test circuit is masking real part variation through compliance
  • Single-use assemblies require a fresh sample per trial — avoid re-pressurising a fatigued specimen
Electrical and electronic test equipment (impedance analysers, hipot testers, multimeters)
  • Cable and fixture impedance at high frequencies is a major EV source — fixturing must be included in the GR&R study, not treated as infrastructure
  • For production hipot testers, contact resistance at the test probes is the dominant EV driver; verify probe tip condition before every study
  • Bioimpedance or RF device measurements are highly sensitive to grounding and shielding — run the study in the actual production environment, not a lab bench
  • Multimeter GR&R studies on resistance < 1 Ω must use 4-wire Kelvin connections; 2-wire measurements include lead resistance in EV
Optical and vision systems (automated inspection, laser micrometers)
  • Lighting intensity drift over a work shift is a hidden EV source in camera-based systems; verify illuminator warm-up stabilisation time
  • Focus repeatability in telecentric optics degrades with temperature; if EV increases between morning and afternoon study sessions, suspect thermal focus shift
  • Edge detection threshold settings must be locked and version-controlled — a threshold change between studies invalidates cross-study comparisons
  • For laser micrometer studies on transparent or translucent materials (clear tubing, optical fibers), verify the beam does not refract through the part; use appropriate wavelength
Weighing and mass measurement
  • Draught shields are mandatory for balances < 0.1 g resolution; HVAC airflow is the primary EV source without them
  • Electrostatic charge on plastic or powder parts causes reading drift; use ioniser or anti-static plate in the weighing area
  • Calibrated reference weights must bracket the measurement range; a balance calibrated only at mid-range introduces systematic EV at the extremes
  • Zero drift between trials must be corrected; re-zero before every measurement or the prior measurement's residue contaminates EV

Study design considerations for regulated environments (medical device)

Consideration Guidance
Minimum study size 10 parts × 3 operators × 3 trials is the AIAG baseline. For critical or life-sustaining device dimensions, some customers require 10 × 3 × 5.
Part selection Parts must span the full production tolerance range, not be selected for "convenience." FDA investigators check this.
Blind randomisation Operators should not see each other's readings. Parts should be re-coded to prevent memory effects across trials.
Environment Conduct the study under production conditions, not controlled lab conditions, unless the gage is exclusively lab-based.
Re-study triggers Re-study is required after: equipment repair or replacement, significant process change, new operator qualification, calibration interval expiry, or any out-of-tolerance calibration finding.
Design history file For 21 CFR Part 820 compliance, the GR&R study, raw data, and acceptance decision must be retained in the DHF or DMR.

General manufacturing (IATF 16949 / ISO 9001)

Key standards: IATF 16949:2016 · AIAG APQP 3rd Ed. · ISO 9001:2015

Red flags — general manufacturing

Flag Why it matters
%GR&R > 10% for production gages PPAP submissions require ≤10% for SPC-critical dimensions; ≤20% is the absolute floor for incoming inspection.
ndc < 5 SPC control charts built on data with ndc < 5 produce false signals — Cpk analysis is statistically meaningless.
AV > EV Operator technique is dominant; enforce standardised work instructions and consider dedicated fixtures.
Study not conducted under production conditions IATF 16949 auditors look for production-floor temperature, noise, and fixturing — not controlled lab conditions.
PPAP MSA section incomplete Missing GR&R data is a Level 3 PPAP rejection trigger; all measurement tools listed in the Control Plan must have a current GR&R.
Pre-control SPC not established before GR&R SPC must demonstrate process stability before a GR&R study; an unstable process inflates PV and produces false-acceptable GR&R results.

Study design — general manufacturing

Consideration Guidance
Threshold — production gages ≤10% GR&R for dimensions subject to SPC or Cpk reporting.
Threshold — incoming inspection ≤20% acceptable maximum for attribute or variable go/no-go gages.
PPAP MSA requirement Every measurement tool in the Control Plan must have an active GR&R study. PPAP Level 3 requires submission of the full raw data and MSA summary.
Linearity & bias Required beyond crossed GR&R for any gage used across multiple nominal values or on a production line with significant part range.
ndc minimum ndc ≥ 5 per AIAG; ndc ≥ 8 recommended for SPC-critical parameters.
Environment Conduct under production conditions; temperature, vibration, and fixturing must match the measurement station.

By instrument type — general manufacturing

  • CMM — fixture repeatability must be validated independently; program version must be version-controlled and frozen before study
  • Torque tools — rate of loading is the primary AV source; standardise dwell time and approach direction
  • Load cells — include hysteresis validation; always approach target load from the same direction
  • Vision systems — lock illuminator warm-up time, threshold settings, and calibration block before study
  • Optical comparators — operator edge alignment is the dominant AV source; use crosshair fixtures where possible
  • Thread gauges — go/no-go attribute gages require attribute agreement analysis (kappa), not variable GR&R
  • Surface plates — flatness calibration certificate and temperature soak time must be documented in the study record

Aerospace (AS9100D / NADCAP)

Key standards: AS9100D · NADCAP AC7130 · SAE ARP9013

Red flags — aerospace

Flag Why it matters
%GR&R > 10% for flight-critical dimensions AS9100D and most NADCAP-accredited laboratories mandate ≤10% for safety-critical features — the AIAG ≤30% threshold is not acceptable.
Missing NIST traceability chain NADCAP audits require a documented calibration chain to NIST (or equivalent NMI) for every measurement tool used in a GR&R study.
Gage changed mid-study Frozen process control requirements prohibit hardware or software changes after a NADCAP study begins; any change invalidates the study.
Uncertainty budget not documented ISO/IEC 17025 accreditation requires a formal measurement uncertainty budget (sources, distributions, combined u_c, expanded U = k·u_c).
Study not in production environment For NADCAP AC7130 special processes, gages must be evaluated in the actual process environment (heat treat, NDT, chemical processing bay).
Fixture variation not separated Fixture repeatability must be quantified and either removed or included as a separate variance component.

Study design — aerospace

Consideration Guidance
GR&R threshold ≤10% mandatory for flight-critical dimensions; document customer approval for any marginal result before production release.
Traceability Full calibration chain to NIST (or DIN/NPL/BIPM equivalent) documented; certificates retained with the GR&R record.
Measurement uncertainty budget Per ISO/IEC 17025: identify all u_i sources, assign probability distributions, compute combined u_c and expanded U = k·u_c (k=2, 95% confidence).
Frozen process control No gage changes, probe changes, or program edits between study start and acceptance sign-off. Document the frozen configuration.
Study size Minimum 10 parts × 3 operators × 3 trials; NADCAP routinely requires 5 operators for special process NDT studies.
Part selection Must span the full blueprint tolerance, including runout and form tolerances for complex geometries.

By instrument type — aerospace

  • CMM — temperature-compensated measurements required; document part soak time; probe qualification logged per trial
  • Laser tracker — atmospheric compensation (temperature, pressure, humidity) must be active and logged; retro-target position repeatability is a primary EV source
  • Optical profilometer — traversal direction locked to machining lay direction; cut-off wavelength (λc) fixed and documented
  • Ultrasonic thickness gauge — couplant type, probe pressure, and surface preparation must be standardised; velocity calibration at study temperature
  • Hardness tester (Rockwell/Vickers) — indentation location randomised; reference block calibrated per ASTM E18; material transfer to indenter documented
  • Torque wrench (mil-spec) — calibrated per MIL-DTL-28778; rate of loading standardised; breakaway vs. prevailing torque protocol specified

Automotive (AIAG APQP / PPAP / VDA)

Key standards: AIAG MSA 4th Ed. · AIAG PPAP 4th Ed. · VDA Volume 5

Red flags — automotive

Flag Why it matters
%GR&R > 10% for SPC dimensions PPAP requires ≤10% for dimensions on the Control Plan with SPC; OEM customer portals (GM, Ford, Stellantis) auto-reject submissions above this threshold.
ndc < 5 (AIAG) or < 8 (VDA) AIAG minimum is ndc ≥ 5; VDA Volume 5 recommends ndc ≥ 8 for SPC-critical parameters — document which standard governs.
No linearity and bias study for PPAP Level 3 PPAP Level 3 MSA submissions require linearity and bias studies in addition to crossed GR&R for any gage used across a range of nominal values.
Fixture effect not quantified Body-in-white and assembly gages have significant fixture-induced variation; failing to separate it inflates EV and produces a falsely marginal result.
VDA discrimination ratio < 4 VDA Q_MS = σ_part / σ_GRR ≥ 4 is required (equivalent to ndc ≥ 5 in AIAG terms) — verify which ratio the customer requires.
Study on non-production parts Parts sourced from a prototype or pre-production run do not satisfy PPAP; production-intent parts are required.

Study design — automotive

Consideration Guidance
GR&R threshold ≤10% for PPAP SPC dimensions; ≤20% for lower-risk features with documented customer concurrence.
Linearity and bias Required for PPAP Level 3; fit regression line (bias = a + b·reference); accept if
ndc requirement ≥5 per AIAG; ≥8 per VDA Volume 5 for SPC-critical parameters. State which requirement applies in the MSA record.
PPAP MSA submission Raw data, summary sheet, range chart, and X-bar chart are required in the PPAP package; Control Plan must reference the gage system ID.
Study timing Must be performed after production tooling is installed and capable; PPAP cannot be submitted on pre-production data.
Re-study frequency Annual re-study is the IATF 16949 baseline; OEMs may specify more frequent intervals for critical dimensions.

By instrument type — automotive

  • Air gauge — temperature of supply air and part must be stabilised; calibration master rings must have documented temperature coefficients
  • CMM — production-environment CMM preferred over metrology lab CMM for PPAP; temperature compensation logged
  • Profilometer — traversal speed, force, and cut-off wavelength locked; Ra vs. Rz vs. Rq parameter must match drawing callout
  • Hardness tester — indentation randomisation required; Brinell ball condition documented; test block calibration within 90 days
  • Vision system — illumination intensity, gain, and edge threshold settings version-controlled and frozen before study
  • Functional test fixtures — fixture wear is a primary EV source; fixture re-qualification interval must be part of the control plan

NIST / National metrology (ISO/IEC 17025 / VIM)

Key standards: NIST Handbook 44 · VIM (JCGM 200:2012) · ISO/IEC 17025:2017 · ILAC P14

In accredited calibration laboratories, GR&R variance components map directly to Type A measurement uncertainty — but the preferred language is measurement uncertainty (MU), not %GR&R.

Mapping GR&R to uncertainty budget

GR&R Component Uncertainty equivalent Distribution
EV (Repeatability) Type A, within-lab Normal
AV (Reproducibility) Type A, between-operator Normal
PV (Part Variation) Not an uncertainty source — it is the measurand variability
Calibration uncertainty of reference Type B Normal or rectangular
Thermal expansion coefficient × ΔT Type B Rectangular
Resolution / least count Type B Rectangular (u = resolution / 2√3)

Uncertainty budget table structure (ISO/IEC 17025)

Source (x_i) Value Distribution Divisor u_i ν_i
Repeatability (Type A) Normal 1 s̄/√n n−1
Reproducibility (Type A) s_between Normal 1 s_between
Reference std. uncertainty u_ref Normal 1 u_ref
Thermal expansion α·L·ΔT Rectangular √3
Resolution res Rectangular 2√3
Combined u_c √Σu_i² Welch-Satterthwaite
Expanded U (k=2) 2·u_c 95% confidence

CMC (Calibration and Measurement Capability)

The CMC line in a laboratory's scope of accreditation defines the smallest uncertainty achievable under routine conditions. For a GR&R study to be valid at a metrology lab, the expanded uncertainty U of the reference standard must be ≤ 25% of the tolerance being measured (ILAC P14 rule; some customers require ≤ 10%).

Red flags — national metrology

Flag Why it matters
Expanded U not reported ISO/IEC 17025 requires all calibration certificates to state U at k=2; a bare %GR&R number is insufficient for accreditation records.
k factor not stated Coverage factor k must be declared alongside U; different labs use different k values; cross-lab comparisons are invalid without it.
Degrees of freedom not tracked Welch-Satterthwaite effective ν must be computed to validate that k=2 achieves the claimed 95% confidence; ν < 10 requires k > 2.
Reference standard not in scope The reference used for the study must appear in the laboratory's current accreditation scope; expired or out-of-scope use is a nonconformity.
Environmental conditions not logged Temperature, humidity, and barometric pressure must be recorded for every measurement in a 17025 study.
No proficiency testing participation ISO/IEC 17025 requires participation in interlaboratory comparisons (PT) for all claimed scopes; a laboratory with no PT record cannot demonstrate metrological traceability.

By instrument type — national metrology

  • Dead-weight tester — piston-cylinder area documented with thermal expansion correction; air buoyancy correction required above 10 kPa
  • Gage blocks — wringing film thickness (≈ 0.01 µm) is a systematic EV source; gage block material must match workpiece CTE for temperature compensation
  • Laser interferometer — refractive index of air correction (Edlén formula) required; vibration isolation and thermal enclosure mandatory
  • Precision balance — magnetic susceptibility of weight set vs. mass piece must be matched; electrostatic discharge from plastic parts is the primary EV source below 1 mg
  • Reference thermometer — self-heating current must be characterised and corrected; calibrated at the exact immersion depth used in the study

Healthcare / clinical laboratory (CLIA / ISO 15189)

Key standards: CLIA 42 CFR Part 493 · ISO 15189:2022 · CLSI EP05-A3 · CLSI EP15-A3

In clinical laboratories, the equivalent of an operator is an analyst, and the equivalent of a trial run is a run (or replicate). Precision is decomposed into within-run, between-run, and between-day components — which map directly to EV, AV, and a third temporal component not captured by standard AIAG GR&R.

Mapping AIAG GR&R to clinical precision

AIAG Component Clinical equivalent CLSI term
EV (within-operator) Within-run imprecision CV_r
AV (between-operator) Between-analyst imprecision
Part Variation (PV) Biological variation between patients CV_bio
(not in AIAG) Between-run imprecision CV_rr
(not in AIAG) Between-day imprecision CV_d

Allowable total error and sigma metric

  • Allowable Total Error (TEa) is the clinical equivalent of engineering tolerance. CLIA proficiency testing criteria define TEa for regulated analytes (e.g., glucose ±10 mg/dL or ±10%, whichever is greater).
  • Sigma metric = (TEa − |bias|) / CV_total — the clinical equivalent of Cpk. Sigma ≥ 6 is world-class; ≥ 4 is acceptable; < 3 requires immediate corrective action.
  • %Imprecision (%CV) replaces %GR&R as the primary metric: CV = (SD / mean) × 100%.

Red flags — clinical laboratory

Flag Why it matters
CV > TEa/4 Imprecision alone consumes more than 25% of the total error budget, leaving insufficient margin for bias.
Sigma metric < 3 Analytical process is unreliable for patient decision-making regardless of QC performance.
Bias not evaluated CLIA and ISO 15189 require both precision and bias (trueness) characterisation; GR&R alone is insufficient.
Matrix not matched Clinical samples have complex matrices; precision studies must use patient-like material (commutable matrix), not pure aqueous standards.
No inter-run replication Within-run CV underestimates total imprecision; between-run and between-day components are required for a complete precision profile (CLSI EP05-A3).
Frozen samples refrozen multiple times Repeated freeze-thaw cycles increase analyte degradation, inflating apparent imprecision.

Study design — clinical laboratory

Consideration Guidance
Precision levels Evaluate at least 2 concentrations: one near the clinical decision point and one at the medical decision limit.
Replicate structure CLSI EP05-A3 recommends 2 replicates × 2 runs/day × 20 days for a complete precision profile.
Analyst rotation Rotate all analysts across all runs; confounded analyst-by-run designs produce biased AV estimates.
Reference materials Use commutable, matrix-matched reference materials (JCTLM-listed where available) for bias evaluation.
TEa definition Use CLIA criteria for regulated analytes; use biological variation-based criteria (RiliBÄK, RCPA) for unregulated tests.
Sigma metric reporting Report sigma metric alongside %CV and bias in the validation summary; regulatory bodies increasingly expect it.

By instrument type — clinical laboratory

  • Pipette — gravimetric verification required; tip lot and aspiration speed must be standardised across analysts; static electricity on polypropylene tips is a primary EV source for volumes < 10 µL
  • Analytical balance — ioniser mandatory below 100 mg; calibration with OIML E2/F1 weights; level bubble checked before each study session
  • Clinical analyzer — reagent lot must be documented; calibration before study start; QC must pass before each run; maintenance log reviewed
  • Glucometer — strip lot number is a dominant EV source; CLIA waiver tests require comparisons across multiple strip lots; capillary vs. venous sample is a systematic bias source
  • Spectrophotometer — wavelength accuracy and stray light verified; cuvette path length uniformity validated; sample temperature thermostatted
  • Centrifuge — speed (RPM) and time verified with tachometer; rotor balance documented; tube type must match serum/plasma separator requirement

Energy / utilities (ISO 50001 / IEC 61869)

Key standards: ISO 50001:2018 · IEC 61869 (instrument transformers) · ASME PTC 19.1 · OIML R 46

Meter accuracy classes

Revenue meters are classified by accuracy class defining maximum permissible error (MPE):

Class MPE (active energy) Typical application
0.1 ±0.1% Reference / check metering
0.2 ±0.2% Revenue metering, transmission
0.5 ±0.5% Revenue metering, distribution
1.0 ±1.0% Industrial sub-metering
2.0 ±2.0% Residential metering

A GR&R-equivalent study for revenue meters is called a Measurement and Verification (M&V) protocol (IPMVP Option D). The %GR&R threshold maps to the M&V measurement uncertainty budget: total uncertainty ≤ 2% of measured energy is the IPMVP baseline.

Red flags — energy / utilities

Flag Why it matters
Instrument transformer ratio error not included Current and voltage transformers introduce ratio and phase errors that directly enter the meter reading; their accuracy class must be convolved with the meter's own uncertainty.
M&V baseline period not representative An M&V protocol built on atypical operating conditions (partial load, weather outlier) produces invalid energy savings estimates.
Flow meter not calibrated at operating conditions Viscosity, density, and Reynolds number at operating temperature affect the meter factor (K-factor); lab calibration at 20°C does not transfer to process conditions.
Grounding and shielding not verified High-voltage transient interference is the dominant EV source for revenue meters in substations; EMI shielding must be verified as part of the study.
Burden of current transformer exceeded Operating a CT above its rated burden degrades accuracy class; verify connected burden before the study.
No reference meter for comparison A GR&R study on a single meter without a certified reference meter cannot detect systematic bias; a transfer standard is required.

Study design — energy / utilities

Consideration Guidance
Study design Three-operator equivalent: morning shift / afternoon shift / night shift for time-of-use meters; each "operator" represents a temporal operating condition.
Environmental conditions Temperature and load profile must be representative of the billing period; summer peak-load and winter base-load are separate study conditions.
Reference standard Calibrated reference meter at accuracy class ≥ 0.2% traceable to NIST or equivalent NMI; or portable reference standard at class 0.05.
Measurement uncertainty Express results as expanded uncertainty U (k=2) per ASME PTC 19.1; %GR&R and MU are equivalent for billing compliance purposes.
Regulatory threshold OIML R 46 requires meters remain within accuracy class MPE throughout the calibration interval; any exceedance is a non-conformance to the billing authority.
M&V protocol Document baseline conditions, measurement boundary, and sampling frequency per IPMVP Annex B.

By instrument type — energy / utilities

  • Revenue meter — sealed, tamper-evident; accuracy class and serial number documented; time-sync via GPS or NTP required for interval metering studies
  • Current transformer — ratio error and phase displacement tested at 5%, 20%, 100%, and 120% of rated current; class P vs. class PX requirements differ by application
  • Flow meter (turbine/ultrasonic/Coriolis) — K-factor calibration at operating viscosity; installation effects (upstream straight pipe) must be replicated in study
  • Pressure transmitter — static pressure effect and temperature coefficient documented; two-point calibration at operating range endpoints
  • Thermocouple — cold-junction compensation validated; extension cable EMF characterised; immersion depth effect tested at study temperature
  • Power analyzer — crest factor rating verified against load waveform; harmonic analysis to confirm no spectral aliasing at sampling rate used

Defense / government (MIL-STD / DCSA)

Key standards: MIL-STD-45662A · MIL-HDBK-1828 · ANSI/NCSL Z540.3

Test Uncertainty Ratio (TUR)

The Defense/government metrology community uses TUR (Test Uncertainty Ratio) rather than %GR&R as the primary acceptance criterion:

TUR = Tolerance / (2 × U_measurement) where U_measurement is the expanded uncertainty (k=2) of the measurement system.

  • TUR ≥ 4:1 — ANSI/NCSL Z540.3 default requirement; no guard band required
  • TUR 3:1 to 4:1 — guard banding required; acceptance limit set inside tolerance to compensate for measurement uncertainty
  • TUR < 3:1 — special approval required; document risk acceptance in the calibration record

The %GR&R maps to TUR as: TUR ≈ Tolerance / (5.15 × σ_GRR)

Guard banding

When TUR < 4:1, the acceptance limit is moved inside the tolerance by the guard band width:

  • Guard band = k_GB × U_measurement, where k_GB is chosen to achieve the required false-accept probability
  • ANSI/NCSL Z540.3 default: k_GB = 1.0 (one-sided guard band of U_measurement)
  • For safety-critical parameters, some programs require k_GB = 2.0

Red flags — defense / government

Flag Why it matters
TUR < 4:1 without guard banding Direct Z540.3 violation; product accepted within the measurement uncertainty zone may be non-conforming.
Calibration certificates missing uncertainty statements MIL-STD-45662A and Z540.3 require uncertainty on all calibration certificates; a certificate without uncertainty is non-compliant.
Classified hardware study without access log DCSA inspection requires a documented chain-of-custody and access log for GR&R studies on classified hardware; unlogged access is a security violation.
Measuring and test equipment not on approved list Programs with Government-Furnished Equipment (GFE) requirements must use only approved M&TE; substitutions require Engineering Change Proposal (ECP).
Calibration interval exceeded Any M&TE used past its calibration due date invalidates all measurements taken after the last valid calibration.
Environmental monitoring not recorded MIL-HDBK-1828 requires temperature, humidity, and vibration monitoring during precision metrology studies.

Study design — defense / government

Consideration Guidance
TUR requirement TUR ≥ 4:1 per ANSI/NCSL Z540.3; document TUR in the calibration record alongside %GR&R.
Guard banding Apply guard band when TUR < 4:1; document guard band width, k_GB value, and false-accept risk in the test procedure.
Classified hardware Controlled area required; access log with date/time/personnel for every measurement session.
Documentation Calibration record must include: serial number, date, result, uncertainty, TUR, standard used, environmental conditions, technician ID.
Traceability NIST traceability is mandatory; for overseas operations, host-nation NMI traceability is acceptable when documented in the calibration procedure.
Interval analysis MIL-HDBK-1828 Method 3 (reliability-based interval) is preferred over fixed intervals; recalibration interval extended when in-tolerance rate > 95%.

By instrument type — defense / government

  • Torque wrench (mil-spec) — calibrated per MIL-DTL-28778; clockwise and counter-clockwise calibrated separately; storage in torque-relief position required
  • Hardness tester — per MIL-STD-1261; reference test block verified against NIST SRM before each study session
  • CMM — per ASME B89.4.10348; volumetric performance verified with ball-bar or step gauge; temperature compensation logged
  • Pressure gauge — deadweight calibration or NIST-traceable piston gauge; hysteresis tested in both directions across full range
  • Electrical multimeter (calibrated) — EMC shielding verified; lead resistance subtracted for resistance < 1 Ω; RF immunity test for field use
  • Laser range finder — atmospheric correction (temperature, pressure) applied; retro-target geometry documented; eye-safety class verified before field study

Regulatory Reference

  • AIAG Measurement Systems Analysis Reference Manual, 4th Edition (2010)
  • 21 CFR Part 820.72 — Inspection, Measuring, and Test Equipment (FDA QSR)
  • ISO 13485:2016 — Medical devices quality management systems
  • IATF 16949:2016 — Automotive quality management systems
  • AS9100D / NADCAP AC7130 — Aerospace quality management and measurement systems
  • VDA Volume 5 — Automotive measurement systems analysis (German OEM supplement)
  • ISO/IEC 17025:2017 — General requirements for calibration and testing laboratories
  • ANSI/NCSL Z540.3 — Requirements for the calibration of measuring and test equipment
  • CLIA 42 CFR Part 493 — Clinical Laboratory Improvement Amendments
  • ISO 15189:2022 — Medical laboratories — Requirements for quality and competence
  • ISO 50001:2018 — Energy management systems
  • MIL-STD-45662A — Calibration system requirements (U.S. DoD)

Reports generated by this tool are quality records. Retain per applicable document control procedures.

About

AIAG MSA 4th Edition Gauge R&R analysis tool for manufacturing quality engineering. Generates PDF reports for 21 CFR 820.72 compliance.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors