Traditionally, problem solving environments like Octave provide simple interfaces to numerical linear algebra, special function evaluation, root finding, and other tools. Special functions (such as Bessel functions, exponential integrals, LambertW, etc) are expected by users to "just work". But many of Octave's special functions could be improved to improve their numerical accuracy. Generally a user might expect these to be accurate to full 15 digits. Software testing is important to Octave; this project would improve the tests of many special functions, in particular by comparing the output with slow-but-accurate symbolic computations.