Pairwise test case generator

Why pairwise testing works

Many defects are triggered by one setting or by the interaction of two. Covering every pair of values catches those without running every combination. Four parameters with three values each have 81 combinations; this generator covers every pair in 9 test cases, which is the smallest possible number for that model.

Choose 3-way coverage for risky areas where three settings might interact. It needs more test cases, but still far fewer than all combinations.

Pairwise testing chooses which combinations to run. You still decide the values, so pick them with equivalence partitions and boundary values: see how to write effective test cases.

To see what a generated set catches on a real app, and what it misses, read pairwise testing on Sauce Demo, measured against every combination.

What makes this generator different

• Rules that survive edits. Rules are text, so changing a value never deletes them. A rule that names a value you removed is reported with its line number.
• Other names for a value. Write Chrome | Google Chrome and cases, rules, or old tables that say either name still match. Exports always use the first name. When an old case differs only in spacing or punctuation, like IE11, the check suggests the alias to add.
• Invalid values. Values starting with ~ appear at most once per test case, so a failing negative test points at one bad input.
• Required cases. Start from test cases you already run; the generator only adds what they don't cover.
• Check the cases you already have. Paste them as CSV or from a spreadsheet to see which pairs they cover, which are missing, and which rows break a rule or repeat another. Then keep them and add only the missing cases. After a release adds a parameter or removes a value, last release's cases still load: each affected case is listed for review, and filling chooses only the values it lacks. Cases keep their IDs, changed cases are marked for review, and you can export the updated set with IDs or copy a change summary. Coverage is counted against your model, so it can't tell you about options the model leaves out.
• Checked coverage. The summary counts every pair or triple a valid test case can contain, and every one of them is covered. Pairs your rules make impossible are listed separately.
• Ready to use. Copy CSV, Markdown, JSON, or a Gherkin Examples table for a Cucumber Scenario Outline, or share a link that keeps the whole model in the URL.
• Saved for next release. “Save matrix” writes the model and its cases to a JSON file on your computer; “Open matrix” reads one back, with the saved cases loaded straight into the check above. Nothing is uploaded and there is no account: the file is yours, and it stays readable by later versions of this tool.

The generator is deterministic: the same model always gives the same cases. For 2-way coverage of models without rules, invalid values, or required cases, it also tries two known constructions and keeps whichever set is smallest: the binary construction of Kleitman, Spencer, and Katona when every parameter has two values (100 such parameters need 10 cases, the proven minimum), and Bush's orthogonal arrays when the values fit a finite field (6 parameters with 5 values need 25). When the summary says no set can be smaller, that has been checked against the known lower bound. Otherwise the result comes from a greedy search, which may be larger than necessary: for 13 parameters with 3 values it finds 19 cases. It runs in your browser, so very large 3-way models can take a few seconds.