So if we do import time *or* test load time parameterisation then we can do separate failure reporting. We may still want to improve "test selection" for parameterised tests.

There are use cases for run time parameterisation (for example generating tests from a database that is only available during the test run was one use case someone has that can't be met by early parameterisation).

If we do run time parameterisation then I think we can only maintain compatibility with existing uses of test suites by having a single test case instance per parameterised test, and single combined failure report.

I would prefer *not* to add run time parameterisation.