Measurement Control
Last updated on October 28, 2026
Yield summarizes the outcome of a Run, while measurement control reaches inside the run to expose every Measurement and every limit at the phase level.
You get two views: Cpk and control charts. For phase-level failure counts and durations, see Phase Pareto.
Process Capability Index (Cpk)
Cpk measures how consistently a measurement stays inside its limits. It applies to numeric measurements with at least one limit, and TofuPilot recomputes it in real time as new runs land.
Formula
The Cpk formula uses the mean and standard deviation of the selected runs.
Cpk = min[(upper_limit - μ) / (3σ), (μ - lower_limit) / (3σ)]Here, μ is the mean across the selected runs and σ is the standard deviation.
Interpretation
You can read Cpk against the same thresholds the industry uses.
| Cpk | Meaning |
|---|---|
| > 1.33 | Well-centered and stable. |
| ≈ 1.00 | Marginal. A small drift will start producing failures. |
| < 1.00 | Not capable. The failure rate is unacceptable. |
Capability table
The Capability table lists every numeric measurement on every phase, sorted by Cpk. Sort ascending to surface the weakest ones first.
Click a row to see Cpk over time. The view shows daily Cpk values for the lower and upper limits separately.
Control charts
A control chart plots a single measurement over time alongside its limits, mean, and ±6σ band, so you can spot drift at a glance.
| Measurement type | Chart |
|---|---|
| Numeric | Values, lower and upper limits, mean, ±6σ band, histogram with a Gaussian overlay. |
| Boolean | Expected values in green. Unexpected values in pink. |
| String | Expected values in green. Unexpected values in pink. When no validators are set, every value renders in gray. |
| Multi-dimensional | Every run's curve overlaid on one chart, or a quantile band over all runs. See Multi-dimensional measurements. |
For numeric measurements, if the Procedure changes limits over time, the chart reflects the history with day-by-day limit curves.
Multi-dimensional measurements
A multi-dimensional measurement records a curve per run, such as a frequency sweep or an efficiency curve. Select it in the measurement filter to compare the curves of every run in the window, for example across revisions or batches.
| Mode | Shows |
|---|---|
| Waveform | One line per run, up to 100 curves: reference units first, then the most recent runs. The band mode covers every run. |
| Scatter | The same samples as points. |
| Band | The median, the p25 to p75 range and the p05 to p95 range at each X value, over every run in the window. No curve limit applies. |
- Series: when the measurement has several Y series, pick the one to display. A second series can be drawn against a right-hand axis.
- Limits: the chart draws the lower and upper limits of the displayed series when its min or max aggregations carry validators.
- Reference units: units marked golden or failing are drawn in their own color, and units selected in the measurement table below are highlighted against the others.
- Point reduction: a crowded chart reduces each curve to fit. The reduction keeps the minimum and maximum of each bucket, so peaks are never smoothed out. The header shows how many curves were reduced.
The band groups samples by exact X value and does not interpolate. When runs use different X grids, a position simply carries fewer runs.
To compare the same curves in your own scripts, read them through the REST API: each run returns the X and Y data of its multi-dimensional measurements.
Filters
Every measurement control view shares the same filter bar.
- Date range: pick a preset or a custom window.
- Source: choose user runs from CLI or production runs from a Station.
- Batch, procedure version, part name, and revision: slice the data the same way you would on the yield view.
- Phase and measurement: drill into one of either to render a control chart or a Cpk trend.
How is this guide?
Run Analytics
Learn how TofuPilot computes first-pass yield from each unit's first attempt, and how retests, aborted runs and filters affect the count.
Phase Pareto
Find the weak step in your test with TofuPilot Phase Pareto: rank phases by failures, retries, and wall-clock duration, then drill into the measurements behind each failure.