Stata is extensively and continually tested for accuracy on all supported platforms (Windows, Mac, Linux, etc.). A suite of tests, including over 1.4 million lines of test script, produces over 3 million lines of output (as of 14 May 2013). This output is electronically compared to previous runs of the test suite and against runs on other platforms. Any differences are examined and resolved by Stata's statisticians and software engineers before a new release of Stata (or any subsequent free updates) is sent to users.
The actual number of tests is even larger than is indicated by the 3 million lines of output. Many of the tests do not produce output but quietly do the computations and compare the results against the correct results. If results are outside accepted tolerance levels, the test fails and produces error messages.
You can view the performance of Stata on the following well-known public test suites: