For the study, the researchers ran 30 tests, 16 in the Python programming language and 14 in JavaScript, that generated 19,200 code samples per test, for a total of 576,000 code samples.