Is the process in control? Common-cause noise vs a real signal
1. Before you start
Every repeating process wobbles. Machine the same bore a thousand times and no two parts come out identical — the measurement drifts up and down inside a band, forever, even when nothing is wrong. The whole skill in this course is telling two kinds of wobble apart. Common-cause variation is the routine, ever-present noise baked into a stable process: dozens of tiny influences (ambient temperature, material lot, operator, gauge) that no single one of which you can name or fix, adding up to a steady band of scatter. Special-cause variation is a signal: a specific, assignable event — a worn tool, a wrong material lot, a mis-set fixture — that pushes the process outside that band or into a pattern it does not normally make.
A tiny example. Suppose a stable machining line has produced bore diameters that, for months, have scattered between 24.94 mm and 25.06 mm around a 25.00 mm target. Today’s part reads 25.05 mm. That is inside the band this process has always made — it is common-cause noise, and adjusting the machine to “correct” it will make the next part worse, not better. But a part reading 25.12 mm is outside everything the process has ever done: that is a signal worth chasing down. Same two numbers, opposite calls — and getting the call wrong is expensive in both directions.
Three honest statements before you start:
- This is a Decide course. You read a situation, learn the ideas, and make a call. You do not write or run any code, and there is no arithmetic beyond reading a number against a band.
- The company below, Meridian Castings, is a composite — an invented plant built from ordinary, illustrative figures so the reasoning is clean. No number here is a claim about any real firm.
- This is not a certification. It proves, to you, that you can tell noise from a signal and defend when to touch a process and when to leave it alone.
2. The Situation
Meridian Castings, a composite aluminium die-casting plant, machines a pump-housing bore to a 25.00 mm target, and the daily scrap rate just ticked from its usual low single digits up to a number the plant manager does not like. He wants the setter to adjust the machine this shift and “get it back on target.” The setter thinks today looks like an ordinary bad day, not a broken process — and if she is right, adjusting will inject more variation, not less. Whoever makes the wrong call pays: intervene on noise and you destabilise a good process; ignore a real shift and you ship bad parts by the pallet.
The trap is that a single ugly number always feels like a problem you should fix right now — and that feeling, acted on, is the most common way a stable process gets made worse.
3. What you’ll be able to do
After this course you will be able to:
- Classify a data point as common-cause noise or a special-cause signal by reading it against the process’s own band and pattern — and say why the raw size of the number does not settle it.
- Explain why reacting to common-cause noise (“tampering”) increases variation, using Deming’s funnel, and name the specific decision that adds the variance.
- Read a control chart and a run rule to separate signal from noise — a point beyond the limits, or a non-random pattern inside them — without a calculator.
- Name the two mistakes — over-reacting to noise and missing a real shift — decide which one a given reaction commits, and defend the call you would make instead.
4. Prerequisites & time box
Prerequisites: none beyond comfort reading a short series of numbers and the idea that a repeating process has natural scatter. No spreadsheet or setup — the Decide hall is read-and-decide in the browser; see the Decide hall’s how-to-read page if this is your first concept course. No prior Decide course is assumed, though the judgment here rhymes with any course about not over-reading a single data point.
Time box: about 24 minutes of reading (measured), plus real thinking time on the call in section 7. That is under the 25-minute cap for a concept course.
Difficulty: 6 / 8 — a manager-level decision: several factors move at once and the naive reading points the wrong way, so you have to reason past it.
Free-tier honesty: no signups, no paid tools, requires_gpu: false. Nothing here costs money
to learn.
5. The case & where the numbers come from
Meridian Castings is a composite manufacturing plant: its target, control band, and the daily measurements below are in-course illustrative assumptions, chosen for clean reasoning and clearly labelled as such — not drawn from or claimed about any real firm. The ideas — common- and special-cause variation, the control chart, run rules, and tampering — are standard statistical process control, developed by Walter Shewhart and W. Edwards Deming, and cited in section 11. No figure here requires arithmetic; every call is made by reading a number against a band or spotting a pattern, so you can reproduce each judgment by eye.
The stable process, as it has behaved for months before today (all figures illustrative):
| Item | Figure (illustrative) |
|---|---|
| Bore diameter target (centre line) | 25.00 mm |
| Lower / upper control limit of the stable process | 24.94 mm / 25.06 mm |
| Typical daily scrap rate when in control | 2%–4% |
| Today’s scrap rate that alarmed the manager | 6% |
Hold the control band, 24.94–25.06 mm. It is not the customer’s tolerance and it is not a target anyone chose; it is the voice of this process — the range of scatter the stable process produces on its own. Everything in section 6 is about deciding, for a given point, whether it belongs to that band or is telling you something new.
6. The Concepts
Common-cause and special-cause variation
A stable, repeating process has a personality: a centre it hovers around and a band it scatters within, produced by many small causes acting together. Common-cause variation is that routine band — the sum of countless minor influences (a slightly warmer morning, a new pallet of aluminium, a different operator, gauge repeatability) where no single cause is worth chasing because none of them, alone, moved the number much. A process that shows only common-cause variation is called in control: not perfect, not free of scatter, but predictable — you know the band tomorrow’s parts will land in.
Special-cause variation is different in kind, not just in size. It is an assignable event — a specific thing that happened — that pushes the process outside its usual band or into a pattern it does not normally make: a tool that chipped, a fixture that slipped, a material lot off-spec, a setter who changed something. A special cause is worth finding precisely because it has a single findable cause; fix that one thing and the process returns to its band.
Read Meridian’s stable history as a series (illustrative daily bore averages, mm):
25.01 · 24.98 · 25.03 · 24.97 · 25.02 · 24.99 · 25.04 · 24.96 · 25.02 · 25.00
Every point sits inside 24.94–25.06, and there is no drift or run — the sequence looks like coin scatter around 25.00. That is a process in control. The judgment call that trips people is that the size of a single number does not, by itself, tell you which kind of variation it is. A 25.05 in that series is a high point, but it is inside the band the process always makes, so it is common cause — noise. To call something a special cause you need evidence it is outside the process’s own voice, not just outside your comfort. Shewhart’s insight, in 1924, was exactly this: draw the band the process makes when nothing special is happening, and judge every new point against that, not against a target or a gut feeling.
Assumptions and limits. The frame assumes the band was estimated from a genuinely stable stretch of the process; a band drawn from a period that already had special causes in it is too wide and hides real signals. It assumes the gauge is trustworthy — a wandering gauge manufactures fake signals. And “in control” is not “good”: a process can be in control yet centred on the wrong target or too wide for the customer’s tolerance. In-control tells you the process is predictable — the precondition for improving it — not that it is acceptable.
Tampering: why reacting to noise makes it worse
The costly instinct is to treat every deviation as a special cause and adjust. Deming called this tampering: reacting to common-cause noise as if it were a signal, and the counter-intuitive result is that it increases variation rather than reducing it.
The demonstration is Deming’s funnel experiment. Drop marbles through a funnel aimed at a target; they scatter around it — common cause. Now “improve” by nudging the funnel after each drop to compensate for where the last marble landed: if the last one fell right, move left. Because each landing was mostly noise, you are steering off of noise, and the scatter gets wider — often substantially — than if you had bolted the funnel down and left it. Every over-correction adds the previous error back into the next shot.
That is what would happen at Meridian if the setter adjusts because today’s part read 25.05. Suppose she nudges the tool down. Tomorrow the process — still just doing its normal scatter — lands low, now plus her adjustment, so it reads lower still; she nudges back up; and the band gets wider than the one it started with. She has taken a stable 24.94–25.06 process and, by responding to each wobble, made it wobble more. The specific decision that adds the variance is adjusting the process in response to a point that was inside its own band. The discipline is blunt: when a process is in control, you do not react to individual points. You leave the setting alone and, if the routine band is too wide, you improve the system — better tooling, tighter fixturing, a more uniform material — which is the only thing that shrinks common-cause variation.
The control chart and run rules
The tool that keeps you from tampering is the control chart (Shewhart chart): the running series plotted against a centre line (the process average) and control limits above and below it, set at the edge of the process’s natural band — conventionally three standard deviations out, which is why a healthy in-control process almost never crosses them by chance. You do not need to compute those limits here; treat Meridian’s 24.94/25.06 as the drawn limits and read points against them.
The chart turns “is this noise or a signal?” into rules you can apply by eye. The first is simple: a single point beyond a control limit is a special cause — investigate it. But signals also hide inside the limits as non-random patterns, and this is where judgment earns its keep. The run rules (the Western Electric and Nelson rules) flag these. The most important:
- A run of 8 (or more) consecutive points on the same side of the centre line is a special cause — even if every point is inside the limits. A fair process should cross its centre line often; eight in a row above it says the centre has shifted, and something assignable moved it.
- A steady trend of 6–7 points climbing or falling signals a drift — a tool wearing, a bath heating up — not random scatter.
Read two new Meridian days against the chart:
- Point A — a 25.05 after the stable run above. Inside the 25.06 limit, no run, no trend. It is the high edge of ordinary scatter: common cause, leave it alone.
- Point B — the series turns to
25.01 · 25.02 · 25.03 · 25.02 · 25.04 · 25.03 · 25.05 · 25.04. Every point is inside the limits, and no single one looks alarming — yet all eight sit above the 25.00 centre line. By the run-of-8 rule that is a special cause: the process centre has shifted up, and there is an assignable reason to find, even though no point broke a limit.
Point B is the whole point of a control chart: the naked eye, scanning for a scary number, sees nothing wrong; the run rule sees a process that has quietly moved. Signal is not always the biggest number on the page.
The two mistakes: overreacting and underreacting
Deming framed process decisions as guarding against two mistakes, and you cannot drive both to zero at once — reducing one raises the other, so the job is to balance them, not eliminate them.
- Mistake 1 — react to an outcome as a special cause when it came from common causes. This is tampering. You adjust a process that was fine, add variation, and often chase a ghost cause you will never find because there wasn’t one. The tell: you are reacting to a point that is inside the band and shows no pattern.
- Mistake 2 — treat an outcome as common cause when it was in fact special. You miss a real, assignable shift — a worn tool, a bad lot — and let the process keep making bad parts because “it’s just noise.” The tell: a point beyond a limit, or a run or trend, that you waved off.
Meridian’s manager, adjusting because scrap hit 6% on one day inside the normal band, is walking into Mistake 1. But if that 6% were the eighth straight day of climbing scrap, or came with a point outside the bore limits, waving it off as “a bad day” would be Mistake 2. The control chart exists to hold the line between them: it says react only to points beyond the limits or to a non-random pattern, and to nothing else. That single rule is what separates disciplined process control from chasing every wobble — or ignoring every one.
7. Your Call
You have seen how common- and special-cause variation, tampering, the control chart, and the two mistakes decide when Meridian should touch its machining line. Now a different decision lands on your desk.
Riverbend Bottling is a composite beverage-filling plant — a different company and a different manufacturing process from Meridian’s die-casting line. Its filler doses 500 mL bottles, and an in-control line has, for months, produced fill volumes scattered between 496 mL and 504 mL around a 500 mL centre line (illustrative). You run the shift, and there is a hard external constraint the Meridian case did not have: regulators require the average fill to stay at or above the labelled 500 mL, and an underfilled lot must be quarantined and re-inspected — a real cost the moment fill runs light. Every question below is about one filling line reading; make each call on its own.
How this differs from the taught case (the transfer): this is a different company and sector (Riverbend, a beverage-filling plant, not Meridian’s aluminium die-casting), the numbers are different (a 500 mL fill with a 496–504 band, so you must read a new band rather than recall Meridian’s), the decision type differs (a regulated release/quarantine-and-adjust call, not a scrap-rate machine-adjust call), and it adds a constraint the taught case lacked (a legal minimum-fill floor with a quarantine cost). The core concept is the same: telling common-cause noise apart from a special-cause signal before you act.
8. Self-check
Before you write the memo, make sure you can say each of these in one line:
- What is the difference between common-cause and special-cause variation, and why does the size of a single number not settle which one you are looking at?
- Why does adjusting a process in response to a point inside its own band make variation worse, and what is that failure called?
- What are two ways a control chart flags a special cause — one obvious, one that hides inside the limits?
- Which of Deming’s two mistakes is a given reaction committing, and what would you do instead?
If any is fuzzy, reread section 6: common vs special cause, tampering, the control chart and run rules, and the two mistakes are the whole course.
9. Stretch
Push the thinking further on your own:
- Riverbend’s line is in control but its band (496–504 mL) sits close to the 500 mL legal floor. Would raising the centre line to 502 mL fix the compliance risk, and what does it cost? (Think about what moves versus what stays the same when you shift the centre but not the width.)
- A brand-new filler has run for only two days. You have almost no history, so no trustworthy band yet. How should you decide whether a low reading is a signal before a control chart is even possible — and what is the risk of drawing limits from those two days? (The genuinely harder one: what does “in control” even mean before you have established the band?)
- Your gauge itself is suspect — it may read a few tenths high or low at random. Walk through how an untrustworthy gauge would show up on the chart, and why it can create both of Deming’s mistakes at once.
10. Ship it — your decision memo
Write a one-page memo to Riverbend’s operations lead about a specific reading. State the call (for the 503 mL bottle: hold the setting; for the run of eight above centre or the 493 mL point: investigate for an assignable cause). Show the reasoning in two or three lines (which band or rule you read the point against, and whether it is common-cause noise or a special-cause signal). Name what you rejected (adjusting off a single in-limit point) and why (tampering widens the band). Name the one thing that would change your mind (a point beyond a limit, or a run or trend inside them). Keep it to a single page an operations lead grasps in two minutes. This memo is your own argued claim — not a credential.
11. Sources
Meridian Castings and Riverbend Bottling, and every figure attached to them — the targets, control bands, daily measurements, and scrap rates — are composite and illustrative, constructed for clean teaching, not drawn from or claimed about any real company. The ideas used to reason about them are standard statistical process control; references below.
| Concept / claim | Source (publisher) | URL | Accessed |
|---|---|---|---|
| Common- vs special-cause variation, in control, and SPC overall | Wikipedia — Statistical process control | https://en.wikipedia.org/wiki/Statistical_process_control | 2026-07-20 |
| The control chart: centre line, control limits, signal vs noise | Wikipedia — Control chart | https://en.wikipedia.org/wiki/Control_chart | 2026-07-20 |
| Shewhart’s founding of the control-chart method (1924) | Wikipedia — Walter A. Shewhart | https://en.wikipedia.org/wiki/Walter_A._Shewhart | 2026-07-20 |
| Deming, tampering, the funnel experiment, and the two mistakes | Wikipedia — W. Edwards Deming | https://en.wikipedia.org/wiki/W._Edwards_Deming | 2026-07-20 |
| The run-of-8 and trend run rules for signals inside the limits | Wikipedia — Western Electric rules | https://en.wikipedia.org/wiki/Western_Electric_rules | 2026-07-20 |
| Additional run-rule set for detecting non-random patterns | Wikipedia — Nelson rules | https://en.wikipedia.org/wiki/Nelson_rules | 2026-07-20 |
Next up
That’s the top of the Operations & Process track. Browse all courses → to pick your next call.