Software checks use recorded inputs and outputs
The wording-checker test note records twelve runs of rules version 1.0. Each row gives the input, the signals returned and a comment on the result. The set includes useful alerts, false alarms and missed problems.
Download its CSV, paste one input into the checker and compare the output. These are selected test cases, so the note does not estimate accuracy on all surveys. A change to the rules needs a new dated test record.
Worked calculations keep the numbers visible
The rating-scale example starts with twenty fictional responses, including four “cannot judge” answers. It shows why the favorable share is 40% of all responses but 50% of the sixteen valid ratings. The downloadable counts let you repeat the arithmetic.
The decision-brief example follows a made-up test from planning assumptions to a decision. Its counts, formulas and thresholds are supplied together. They teach the calculation; they are not evidence that a real message performed well.
Calculators state their assumptions
The margin-of-error tool models sampling uncertainty under the assumptions shown beside its inputs. The comparison calculator uses independent groups and withholds its normal-approximation inference for sparse counts. The fieldwork planner estimates recruitment and cost; it does not calculate statistical power.
Use the example inputs on each tool page to check a result. Keep the denominator and unit with the number when you copy it into a report. An interval cannot account for an unrepresentative sample or a badly worded question.
Guides separate sources from our applications
Sources sit beside the claims they support and at the end of each guide. Our question rewrites, prompts, worksheets and fictional scenarios apply those methods to a particular task. Citing a methods source does not make an example a validated questionnaire.
For a real study, record the exact questions and response options, participant source, field dates, exclusions, assignment, analysis and limits. The reporting checklist lists those records. None is supplied by a generated example.
Try a small check before relying on the result
- Start with a published example and confirm you can reproduce it.
- Replace it with made-up inputs that resemble your task.
- Read the relevant limitation before applying your own data.
- Keep a copy of the input, output and tool or worksheet version.
If a result does not match, send the page and a non-sensitive example. Material errors are recorded in the correction log. The editorial standards explain AI assistance and review dates.