Measure the observed error rate in a data-labeling audit and project how many errors may exist in the full batch. Provide the audited sample, failed labels, total batch size, and optional correction time. The results show error percentage, estimated batch errors, expected correct labels, and remediation hours for quality-control planning.
The projected counts scale the sample error rate to the full batch. They are planning estimates and depend on the audit sample representing the entire labeling batch.
Do not use the projection if the sample deliberately overrepresents hard cases unless you first apply appropriate sampling weights.
If 24 errors are found in 400 audited labels, the observed error rate is 6.00%. Applied to 50,000 labels, that implies about 3,000 errors, 47,000 correct labels, and 125 correction hours at 2.5 minutes per error.
For a label-level error rate, count the label once. Track defect-level counts separately if multiple defects per label matter.
The calculator allows it mathematically, but an audit normally samples from the batch, so the batch should generally be at least as large as the sample.
No. It uses the observed rate as a point estimate; use the confidence interval calculator when uncertainty bounds are needed.
Use measured average handling time from similar corrections, including lookup and review, rather than an optimistic target.
Audit comparable tasks with consistent guidelines and report both error rates and sample sizes.
| Input | Unit | Role |
|---|---|---|
| Labels audited | labels | User-provided planning input. |
| Errors found | labels | User-provided planning input. |
| Full batch size | labels | User-provided planning input. |
| Minutes to correct one error | min | User-provided planning input. |