Quality control

The chart is drawn against the material's own statistics

/quality-control — control runs, plotted.

A QC material carries a stored mean and standard deviation, established over at least twenty runs, together with the count they were established from. The chart is drawn against those, never against the statistics of the points currently on screen.

A chart fitted to its own data cannot detect a shift.

/quality-control/materials is where those figures live, with each lot's expiry — and a lot within ninety days of expiring can be filtered for, which is the control the page's own empty state had always described.

A Westgard violation blocks; it does not notify

When a run breaches a rule, the run's QC state becomes out_of_control, and the release gate refuses. It is not a message somebody might read.

A notification can be missed and a gate cannot.

A run that has not been evaluated is not_run, which does not block — the absence of a check is not the same fact as a failed one, and treating them alike would stop a laboratory that has simply not got to its controls yet.

The four quality tools

/quality/tat-optimizer, /quality/six-sigma, /quality/inventory-forecast and /quality/csat are one screen with four fills, reached from a tab strip.

Every figure on them is arithmetic over real records, computed in SQL. There is no model, and nothing here is described as AI — a claim of that kind would be untrue and would fail a store review besides.

Satisfaction is the one that needed a source built for it: a laboratory sends a link, the recipient answers on a page that signs nobody in and reveals nothing about any episode, and the tool publishes counts and a mean. Comments are read, never scored.