Control Chart Calculator

Monitor model metrics, detect unusual variation, compare baselines, apply control rules, inspect violations, and export clear charts for reliable machine learning operations in production.

Results summary

Calculated limits, stability status, and detected control-rule signals.
Not calculated
Observations
Center line
UCL
LCL
Sigma
Violations
Within limits
Longest run above
Longest run below
Capability

Violation table

#LabelValueDifferenceZ distanceRuleSeverityInvestigation note
No calculated violations.

Calculated series

#LabelObservedPlottedCenterUCLLCLStatus
Results appear here after calculation.

Calculator settings

1. Data input
Paste CSV, tab-separated values, comma-separated numbers, or one value per line.
Use this to compare a model version, period, class, or segment.
2. Data mapping and preparation
0 uses row number.
3. Control chart selection
4. Center line, baseline, and limits
0 uses all rows.
5. Control rules
6. Rolling analysis and capability
7. Chart presentation and annotations

Formula used

Individuals chart: CL = x̄, UCL = x̄ + Lσ, LCL = x̄ − Lσ.

Moving range: MRᵢ = |xᵢ − xᵢ₋₁|. The calculator estimates short-term sigma from moving ranges when appropriate.

p chart: CL = p̄, limits use √(p̄(1−p̄)/n).

u chart: CL = ū, limits use √(ū/n).

EWMA: zₜ = λxₜ + (1−λ)zₜ₋₁. CUSUM accumulates positive and negative departures from the target.

How to use

  1. Paste measurements, model metrics, counts, or rates into the primary dataset.
  2. Map the relevant columns and choose a compatible chart type.
  3. Select baseline rows, limit settings, rule tests, and display options.
  4. Calculate the chart and investigate every highlighted control signal.
  5. Export the chart, tables, or complete monitoring report.

Example data structure

DateMetric valueGroupSample sizeDefects
2026-01-010.842A10016
2026-01-020.851A10015
2026-01-030.847A10015

Interpretation guide

Control limits describe expected process variation, while specification limits describe acceptable outcomes. A point outside control limits suggests special-cause variation. Runs and trends can reveal smaller systematic changes.

Machine learning teams can track accuracy, latency, drift, failures, residuals, and confidence. A stable chart does not guarantee adequate model quality. It indicates that recent behavior is statistically consistent.

Recalculate limits only after investigating genuine process changes. Removing inconvenient observations can hide operational problems. Keep documented baselines for model and deployment comparisons.

Frequently asked questions

What is a control chart?

It displays a metric over ordered observations with statistically calculated limits. The chart separates routine variation from unusual signals. It supports disciplined monitoring decisions.

Which chart suits single ML measurements?

An Individuals chart is usually appropriate for one value per interval. A Moving Range chart shows consecutive variation. Use both when monitoring isolated production measurements.

When should I use a p chart?

Use a p chart for proportions with known sample sizes. Sample sizes may vary between observations. Examples include daily misclassification or failure proportions.

What is the difference between UCL and USL?

UCL is a statistical control limit derived from process behavior. USL is a business or engineering specification. They answer different operational questions.

Can I use control charts for model drift?

Yes, chart a drift statistic across ordered monitoring windows. Baselines should represent accepted model behavior. Investigate shifts before changing thresholds.

Why are control limits commonly three sigma?

Three-sigma limits balance false alarms and signal detection. Alternative multipliers may be appropriate for specific risks. Document every chosen monitoring policy.

What does a run above the center mean?

A long run suggests a sustained process level shift. It may follow retraining, data changes, or deployment updates. Review corresponding annotations and system logs.

Should outliers be removed?

Only exclude observations with a documented and justified reason. Outliers often contain valuable incident information. Recalculate baselines after completing the investigation.

Can the calculator compare model versions?

Yes, paste a second compatible dataset for chart overlay. Use identical labels and measurement definitions when possible. Differences become easier to review visually.

What minimum dataset size is recommended?

More baseline observations produce more reliable control limits. Twenty to thirty points is a practical starting range. Small datasets require cautious interpretation.

Related Calculators

Anomaly Score PlotTime-Series Anomaly PlotIsolation Forest Decision PlotReconstruction Error HistogramNormal-versus-Anomalous Scatter PlotThreshold Selection Curve

Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.