Control Chart or Run Chart¶
Visual monitoring tool — instantiates Reference-Baseline Deviation Flagging
Plots observations against a centerline, control limits, reference bands, or expected ranges over time to reveal departures as shifts and trends.
A Control Chart or Run Chart instantiates the archetype in the time domain. It plots a stream of observations against a fixed centerline (the baseline) with control limits drawn a set distance out, so that ordinary variation stays inside a band and genuine departures reveal themselves as points beyond the limits or as non-random patterns — runs, trends, and shifts — across time. Its defining idea is that a single point rarely means anything on its own; meaning lives in how the sequence behaves relative to a stable reference. Unlike a delta table, which shows one snapshot per unit, the control chart's whole value is temporal: it separates the signal of a real change from the constant background hum of noise.
Example¶
An emergency department tracks door-to-provider time — how long each arriving patient waits before a clinician first sees them. Every hour the charge nurse takes the five most recent arrivals and plots their mean wait on an X-bar chart whose centerline sits at the department's in-control average, with control limits at roughly ±3 standard errors derived from its own recent variation. Through a normal shift the points wander harmlessly inside the band — that scatter is the ordinary ebb and flow of arrivals, not a problem to chase. Then a run of eight consecutive points climbs above the centerline. No single point breaks a limit, but eight-in-a-row on one side is one of the classic run signals,[n1] and it flags a sustained upward drift — a downstream bed shortage quietly backing patients up in the waiting room. The charge nurse escalates and opens overflow capacity before waits stretch to where a deteriorating patient goes unseen. The chart caught a trend the delta table's snapshot never could.
How it works¶
The chart is built from the process's own behavior. A centerline is set from the in-control baseline; control limits are computed from the natural variation of the data (commonly ±3 sigma), not from a specification or a wish. Points are plotted in time order. Two families of signal are then read: out-of-limit points (a single observation beyond the band) and run signals (patterns that are individually inside the limits but collectively improbable under stable operation — long runs on one side, steady trends, cyclic waves). A run signal is the chart's native drift detector: it names when the underlying baseline has shifted, distinct from a one-off spike.
Tuning parameters¶
- Limit width — ±2 sigma versus ±3 sigma. Tighter limits catch small shifts sooner but raise false alarms; wider limits are calm but slow to react.
- Subgroup size and frequency — larger, more frequent subgroups detect small shifts faster at higher sampling cost; the choice sets the chart's sensitivity floor.
- Run-rule set — how many pattern rules (eight-on-one-side, six-trending, two-of-three near a limit) are active. More rules catch more drift but compound the false-alarm rate.
- Chart type — individuals, X-bar/R, p-chart, and so on, matched to whether data are counts, measurements, or proportions.
- Recentering policy — whether and when to recompute the centerline and limits, which trades responsiveness to real change against stability of the reference.
When it helps, and when it misleads¶
Its strength is that it distinguishes the two things people constantly confuse: common-cause noise (leave it alone) and special-cause signal (act now). By encoding uncertainty directly into the limit band, it stops teams from reacting to every wiggle, and its run rules surface slow drift long before any single point looks alarming.
Its failure mode is over-control, or tampering: adjusting the process in response to points that are only common-cause noise, which provably increases variation — the lesson of Deming's funnel experiment. The chart also assumes the limits reflect a genuinely stable baseline; limits computed during an unstable period bake the instability in, and a chart never recentered will eventually call a new normal "abnormal" forever. The guarding discipline is to compute limits only from an in-control stretch, act on rule signals rather than on individual scary-looking points, and treat a recenter as a governed event, not a casual redraw.
How it implements the components¶
declared_reference_baseline— the centerline is the reference made visual; every point is read as a distance from it.uncertainty_and_noise_annotation— the control limits are an explicit annotation of natural variation, drawing the boundary between noise and signal onto the chart itself.baseline_drift_monitor— the run rules detect sustained shifts and trends, naming when the process has departed from its baseline rather than merely that one point did.
It does NOT implement deviation_fact_record — persisting each flagged departure as a durable record is the Deviation Event Log's job — nor deviation_routing_policy, the disposition of a flag once raised, which belongs to the Deviation Review Queue.
Related¶
- Instantiates: Reference-Baseline Deviation Flagging — the chart is the pattern rendered as a time series with a variation band.
- Consumes: Baseline Version Register supplies the governed centerline and limits, so a recenter is an accountable version change.
- Compare: Control Chart is the variance-reduction twin — the same time-series monitor read as the guardian of an already-reduced spread — while Control Chart on Residuals charts a model's residuals rather than the raw observations this page plots, watching what a fitted model has failed to explain instead of the measurements themselves.
- Sibling mechanisms: Baseline Delta Table · Baseline Version Register · Deviation Event Log · Deviation Review Queue · Exception Flag Rules Engine · Null-Model Residual Report · Reference Range Flag · Rolling Baseline Comparison · Standardized Residual Score
Editorial Notes¶
Form Classification¶
Form family: Monitoring, Sensing & Alerting
Rationale: Plots observations against a centerline, control limits, reference bands, or expected ranges over time to reveal departures as shifts and trends, making its operative form an ongoing sensing arrangement that repeatedly observes state and surfaces changes or alerts.
Independent corroboration: The frozen evidence defines Control Chart or Run Chart as 'Plots observations against a centerline, control limits, reference bands, or expected ranges over time to reveal departures as shifts and trends', so its operative form is Monitoring, Sensing & Alerting.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Statistical process control cohered time-ordered observations against centerlines, limits, and run rules to reveal shifts and trends beyond ordinary noise.
Related originating lineages:
- Engineering & Design — Industrial quality practice institutionalized Shewhart control charts for process monitoring.
- Medicine & Healthcare — Clinical quality improvement adopted run and control charts for service and patient-safety measures.
Review resolution: Statistical process control is primary, with industrial quality and clinical improvement furnishing genuine institutional lineages for control and run charts across domains.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The Western Electric rules — a canonical set of run-pattern tests for control charts (one point beyond 3 sigma; eight consecutive points on one side of the centerline; two of three near a limit; and so on). They let a chart flag a shifted baseline from a non-random pattern of in-limit points, not just from a single out-of-limit one. ↩