PICT

PICT (Pairwise Independent Combinatorial Testing) is Microsoft’s command-line generator of combinatorial tests: from a text model of parameters, values and constraints, it prints a small table of tests in which every pair of values appears at least once [1].

The quality question it answers: which few tests cover every interaction between two inputs? The method, and why pairs catch most interaction faults, is on Combinatorial Testing.

Run for this card with PICT 3.7.4, October 2026.

Install and quick start

brew install pict        # or build from github.com/microsoft/pict; Windows: download pict.exe

A model for a parking ticket machine, machine.txt (constructed example): one parameter per line, then constraints.

Payment:  cash, card, app
Member:   yes, no
Stay:     short, hours, overnight
Language: en, de, fr
Receipt:  paper, email, none

IF [Payment] = "cash" THEN [Receipt] <> "email";

pict machine.txt > machine.tsv     # pairwise (strength 2)
pict machine.txt /o:3              # every triple instead
pict machine.txt /s                # statistics only

Reading the output

The output is tab-separated, one test per row, with a header (first rows shown):

Payment	Member	Stay	Language	Receipt
cash	no	hours	fr	paper
card	yes	overnight	de	none
app	yes	hours	de	email
card	no	hours	en	none
...

pict machine.txt /s prints:

Combinations:   77
Generated tests:13

All combinations of the five parameters number 3 × 2 × 3 × 3 × 3 = 162; 13 tests cover all 77 allowed pairs. No pairwise suite can be smaller than 9, the product of the two largest parameters (3 × 3). The cash-and-email pair is excluded, and no row contains it. Without the constraint, PICT generated 11 rows; with strength 3 (/o:3), 34.

Common mistakes

• Deleting forbidden rows afterwards instead of writing a constraint. Removing rows from an unconstrained table also removes the other pairs they covered. Put the rule in the model, and PICT covers those pairs elsewhere.
• Expecting PICT to give the expected results. It chooses inputs; each row still needs an oracle, the result the test should get.
• Too many values per parameter. The table grows with the product of the two largest parameters. Use equivalence classes and boundaries (one value per class), not every value.
• Mixing invalid values into normal rows. Mark an invalid value with ~ (Quantity: 1, 5, ~0): PICT then puts at most one invalid value in a row, so one error cannot hide another.
• Assuming the array is right. Check a generated table against the model before relying on it; Combinatorial Testing Tools has a short checking script.

In CI

Keep machine.txt in version control next to the tests and generate the table in the build; PICT gives the same output on every run unless /r asks it to randomise. A parametrised pytest test can read machine.tsv row by row.

Go deeper

• Combinatorial Input Models and Constraints and Invalid Values: building the model.
• Combinatorial Testing Tools: PICT against ACTS and CASA, seed tests (/e), and a pytest file driven by the table.
• ACTS: NIST’s generator, for higher strengths and mixed-strength suites.

---

References

1. Microsoft,
   “PICT: Pairwise Independent Combinatorial Testing, version 3.7.4.” 2022. [Online]. Available: https://github.com/microsoft/pict

---

Disclaimer: AI is used for text summarization, polishing and explaining. Authors have verified all facts and claims. In case of an error, feel free to file an issue.