np Chart: Hold the Gain After the Project Team Walks Away

The np chart monitors the count of defective units in constant-size subgroups for attribute defect monitoring. That is the whole job. You take a fixed number of items each period, count how many of them fail, and plot the count.

It belongs in the Control phase of RDMAICS because Control is where a change stops being a project and starts being the way the work is done. During Improve you proved a new method worked. During Control you need evidence that it keeps working after the team disbands, after the champion moves on, and after the next quarter brings a different priority.

Reach for an np chart when three things are true. The thing you count is pass or fail with no middle ground. Your sample size does not change from period to period. And you care about the number of defectives rather than the proportion. If the sample size moves around, a p chart is the better fit. If you are counting defects rather than defective units, so one item can carry several faults, look at c and u charts instead.

Everything that follows uses the sample data built into the tool. This case is fictional. It exists so you can see the mechanics without needing your own dataset first.

The scenario is a food service operation checking order accuracy. Every day, 100 orders are reviewed. Each order is either correct or defective. There is no partial credit. The count of defective orders is recorded against the day, and each day is tagged with the shift it belongs to, either Lunch shift or Dinner shift. The whole series is labelled Defective orders per 100.

Twelve days of data are loaded. The counts run 6, 7, 5, 8, 6, 7, 15, 5, 6, 8, 7, 6. Read that list quickly and one number stands out. Day 7 recorded 15 defective orders. Every other day sits between 5 and 8. The purpose of the chart is to tell you whether that reaction is justified by the arithmetic or whether your eye is just drawn to the biggest number in a short list.

The input structure is deliberately plain. Four columns carry the data and one setting governs the whole chart.

The observation or period column holds the label for each row, here Day 1 through Day 12. The defective units column holds the count of failing items in that sample. The group or stratum column records Lunch shift or Dinner shift. The series or scenario column lets you keep more than one tracked measure in the same workspace, here just Defective orders per 100. Above the table sits the constant sample size, set to 100.

That sample size field is the one to get right before anything else. An np chart assumes every subgroup contains the same number of items. If you inspected 100 orders on most days but only 60 on one, the chart is no longer valid as built. The limits are calculated on the assumption of a fixed denominator. Change the denominator and the limits stop meaning what they claim to mean.

The tool returns a short block of figures and a plotted chart. Work through the figures in order.

Points is 12, confirming all twelve days were used. Centre is 7.1667, the average count of defective orders across the twelve samples. Upper limit is 14.9047. Lower limit is 0. Sigma is 2.5794. Out of control is 1.

The lower limit of 0 is worth a moment. With a centre near 7 and a sigma near 2.58, three sigma below the centre lands below zero, and a count of defectives cannot be negative. So the lower limit truncates at 0. In practice this means the chart cannot signal an unusually good day through the lower limit alone. Improvement would have to show up as a run of low points under the other Nelson rules.

The signals table lists one entry. Point 7, value 15, triggered rule 1, one point beyond 3 sigma. The tool notes that a single reading this far out is almost never chance. The value 15 sits above the upper limit of 14.9047, so the signal is marginal by arithmetic but real by rule. The tool also offers an AI review that explains the result in plain language if you want a second read before you act.

A Control phase chart earns its place by changing what people do. Here are the decisions this one supports.

First, it tells you when to investigate and when to leave the process alone. Days with 5, 6, 7 and 8 defectives are all inside the limits. They look different to each other but they are the same process. Chasing an explanation for the difference between a 5 day and an 8 day burns effort and often makes things worse. Day 7 at 15 is a different matter. Something happened that day, and finding out what is a reasonable use of time.

Second, it gives the handover a hard edge. When the project team steps back, the chart becomes the standing agreement about what normal looks like. Centre 7.1667 and upper limit 14.9047 are numbers anyone on shift can check against. No interpretation required.

Third, it supports a baseline decision. The tool notes that limits were calculated from all 12 points, and that you can set a baseline above to judge later data against an earlier period. That is the mechanism for holding a gain. Lock the limits to the post-improvement period, then let new days plot against those fixed limits. Drift back towards the old performance shows up as points climbing, not as limits quietly widening to accommodate the slip.

Most np chart problems are set-up problems, not interpretation problems.

The first is a sample size that is not actually constant. Teams often intend to inspect 100 items and then inspect 94 on a busy day. If that happens, either correct the sample or move to a p chart.

The second is recalculating limits every time new data arrives. If the limits move with the data, a slow decline in performance pulls the centre up with it and nothing ever signals. This is the single most common way a Control chart stops controlling anything. Set the baseline and leave it.

The third is reacting to every movement. Eleven of the twelve days here carry no signal. They are the voice of the process, not a series of events needing explanation.

The fourth is ignoring the stratum column. The data carries Lunch shift and Dinner shift tags. The chart as calculated pools them. If you suspect the two shifts behave differently, that is a question to ask deliberately rather than something the pooled chart will answer for you.

Be clear about the boundaries before you rely on this chart in a handover document.

It will not tell you why Day 7 recorded 15 defective orders. Rule 1 fired. The tool says a single reading this far out is almost never chance. That is a statement about probability, not about cause. The cause has to be found by going and looking.

It will not tell you whether 7.1667 defectives per 100 is acceptable. The centre line describes what the process does, not what the customer needs or what the business can afford. A stable process can be stably poor.

It will not detect a problem that arrives in the gaps between samples. You inspect 100 orders. The chart knows nothing about the others.

It will not judge Lunch shift against Dinner shift. The strata are recorded in the data, but the result shown pools all twelve points into one centre and one set of limits.

It will not keep itself current. Twelve points produced these limits. The chart only continues to protect the gain if someone keeps plotting, keeps the baseline fixed, and acts when a rule fires. All eight Nelson rules are applied to the main chart, so the detection is thorough, but detection without response is just record keeping.