Test all lines on signal with a maximally biased sample
!216 (merged) adds a test on minimum bias (which should ideally select nothing for a reasonably sized sample). It's even more interesting to test with signal, however it is not obvious how to do it hundreds of lines/modes.
A practical way to add a (quick) test that is sensitive to many changes is to craft a "maximally" biased sample that contains as many signals as possible and fires as many lines as possible. We can then run HLT with all lines on it. "Meaningful" validation will be rather non-trivial, as we need to correlate and count the decisions with the signal type. In any case, the changes in the output will be useful for detecting potential issues for corner cases (e.g. a regression in efficiency for very high momentum tracks).
Such a sample needs to be prepared whenever we change tags and needs to be updated as new signal samples are available.