The idea is that when-ever we find anything passing validation
where it shouldn't, we add an entry.

To make this somewhat sane, a custom regression-language was
created where you can target the API in a simple human-readable
way. The drawback is that it can be difficult to see what the
real calls are that are being made. This is a balance between
being able to motivate people to write regressions and being
very verbose in case the regression fails.