Every rule. Every interval. Every refusal.
The chart and the capability panel are one surface, computed backend-side so there is no client-side math drift — and built to withhold a number rather than invent one.
The AI never touches your math.
Control charts, capability, Gage R&R and every test are deterministic engines. The AI explains and proposes — it cannot compute a number into existence. Where narration is switched on, the engine authors every figure and every verdict, and any sentence that tries to state one is discarded before you see it.
Before you fix the process, trust the gauge.
Measurement-system error masquerades as process variation, and teams spend months fixing processes that were fine. The gauge study is not an appendix here — it gates what the evidence is allowed to claim.
Gage R&R by the ANOVA method
Crossed studies with variance partitioned into repeatability, reproducibility and part-to-part — not the older average-and-range shortcut. You get %Contribution, %StudyVar, %Tolerance and ndc, banded acceptable / marginal / unacceptable, with thin-design and unbalanced-study flags that bound how strong the rating is allowed to sound. Calibration records sit alongside, with an overdue view.
The trust gate
The gauge verdict follows the data into the analysis: evidence from an untrustworthy gauge is downgraded before conclusions are drawn, and the investigation says so out loud. That is the difference between a measurement study you file and one that changes the answer.
Ask in plain language. Get computed answers.
Your quality engineers should not need to remember which tool answers which question. Ask SPC turns the question into the right statistical call — and the engine, not the model, produces the number.
Program what to watch. The platform watches it.
Between incidents, nobody is looking. A Watchtower routine is a standing watch over a parameter, a machine or a whole line — running on your schedule, with the checks you chose, reporting quietly when there is nothing to say.
It doesn't just alert. It designs the next analysis.
An alert hands you a problem. A proposal hands you the next move — a specific study, with its design pre-declared before any number is computed.
Some problems are only visible from above.
A parameter can be in control on its own chart and badly out of place next to the others. Joint studies look at the machine, the line, and the fleet — with the discipline to not over-read what they find.
Equipment and line studies
Analyse a machine or a whole line as one system: many parameters compared over the same frozen window, plus a multivariate Hotelling T² view that catches the joint excursion no single chart shows — each parameter still inside its own limits while the combination is not.
Cross-equipment common-cause screening
When several machines wobble in the same week, the engine screens for a shared cause and reports what it can support. Temporal coincidence alone is never called a common cause — a co-occurrence is reported as a co-occurrence, and the phrasing does not get upgraded because it would read better.