Start with the inspection question
Count inspected units and units that failed the same criterion. If one part has three scratches, it is still one defective unit here. Counts of individual defects belong to a defect-count model; defect opportunities and DPMO answer another question. Do not combine different products, inspection rules or periods with a changing process just to obtain a narrower interval.
Editable Word sample files
Inspect two free fixed Word samples with synthetic data, methods and review notes. Custom Word exports are not available for purchase. Print / Save PDF remains free.
What the interval means
The observed proportion is k/n. The Wilson score interval without continuity correction describes uncertainty in a binomial process proportion under independent, representative observations with a common probability. Its nominal confidence level describes the procedure over repeated samples, not a probability assigned to this particular fixed parameter. The interval is approximate; coverage can differ from the nominal level, especially with very small counts. This tool does not compute an exact Clopper–Pearson interval.
Method you can reproduce
Let p̂=k/n and let z be the standard-normal quantile for the chosen confidence and direction. The Wilson endpoints are [p̂ + z²/(2n) ± z√(p̂(1−p̂)/n + z²/(4n²))] / [1 + z²/n]. A two-sided interval uses z at (1+confidence)/2. A one-sided bound uses z at confidence; the other endpoint is the physical 0 or 1 boundary, not a separately estimated confidence bound. Results and CSV include the actual z used.
Zero is not proof of zero risk
With zero defectives in 100 independent inspections, a 95% one-sided Wilson upper bound is about 2.63%. A larger random sample narrows the uncertainty, but cannot repair biased selection or an inconsistent measurement system. Do not convert this interval into a guaranteed shipment defect rate.
Choose the next workflow
For changes over time, use attribute control charts. For a predefined lot inspection decision, use an appropriate acceptance sampling plan and its assumptions. Use Pareto analysis to investigate failure categories; categories may overlap unless your classification prevents it. This interval alone establishes neither process stability nor lot acceptance.
Assumptions and scope
Clustering, repeated checks of the same unit, serial dependence, changing rates and misclassification can invalidate a simple binomial interval. If you inspected the entire finite lot, k/n is its observed proportion; this model-based interval is not finite-lot sampling uncertainty. A large sampling fraction without replacement may require a finite-population method. No control limits, capability indices, automatic pass/fail, causal claim or predicted savings are produced.
Method reference: NIST confidence intervals for the proportion defective. The one-sided 26/200 example reproduces its published Wilson calculation within rounding. The report is computed locally without storing counts or placing them in a share URL.
Frequently asked questions
Do I enter defects or defective units?
Defective units. Count each inspected unit once, even if it has multiple defects.
Does zero defective mean the process is perfect?
No. The positive upper confidence bound shows remaining uncertainty under the model.
Can this accept a lot?
No. Lot acceptance needs an appropriate pre-specified sampling plan and operating risks.
Is this an exact interval?
No. It is the Wilson score interval. It is not Clopper–Pearson, adjusted Wald or a control-chart limit.
Are my counts saved?
No. This calculator does not store or upload counts, and does not include them in a URL. Downloaded reports remain under your control.