Compelled as I am to audit everything, I had to compare running bjam in the lib/<x>/test directory with run.py. I'm confident that they do produce (essentially) the same compile, link, and execute commands. Notes, from a regression-runner's POV: * They use different directories: run.py uses {boost-root}/../results/bin.v2/..., and bjam uses {boost-root}/bin.v2/... * run.py sets up bjam with a lot of flags, like --debug-configuration [shouldn't matter for most people] -d2 [important for many, described below] -lxxx [that's lower-case L, with a time-out in seconds: potentially important] preserve-test-targets=off [not sure if this matters or not] * run.py runs from the boost/status directory, I think. Might matter for some tests. Not sure about straight bjam. To make your own script to rebuild/rerun a particular test, run bjam with -d+2 (i.e., -d2) and direct the output to a file. Then copy the compile & link commands out of that file. The -lxxx command (e.g., -l120) matters if a test could hang (e.g., <https://svn.boost.org/trac/boost/ticket/4766>).