Statistical Process Control — the chart everyone draws and half misuse

You plot your measurement over time, then draw two lines — an upper and lower limit — based on how much the process normally bounces:

In plain words: take the long-run average, work out the size of the normal bounce, and draw the lines three bounces above and below. Points inside the lines are ordinary variation — leave the process alone. A point outside the lines is rare enough (about a 1-in-370 event by chance) that something real probably happened — go and look.

That "three bounces" width isn't sacred maths. It's a practical trade-off between two costs: chasing ghosts, and missing real problems.

• Confusing the chart's lines with the customer's needs. The chart's limits describe what your process does. The customer's limits describe what they need. They are different lines, and a process can stay neatly inside its own limits while shipping rubbish. The gap between the two even has its own measure, called $C_{pk}$ — in plain words, "how much room your process has before it starts breaking the customer's rules".
• Reacting to every wiggle. Adjusting a stable process every time a point moves actually makes it more erratic, not less. This is proven, not opinion — it's the famous funnel experiment.
• Charting chaos and calling it monitoring. If the process is unstable, limits drawn from its data are just limits on chaos.

The chart's genius isn't the formula — it's the habit of asking "noise or news?" before reacting. Its weakness is that it only answers that one question. A chart on the wall is surveillance. A chart with an agreed response plan is control.