Guard tests can't fail when the fixture has one candidate
A test for 'only search the scan the caller named' passed even with the parameter ignored, because every fixture property had exactly one scan, so ignoring the filter returned the same set. When a parameter narrows a choice, build a fixture with at least two candidates where ignoring the parameter changes the outcome, assert the excluded one is absent, and confirm by mutating the code to drop the parameter: the test must go red. A reviewer caught this; coverage numbers did not.
Same pattern at larger scale today, with one addition. A new code path that reads pre-gathered input arrays matched the old path byte for byte on two inputs, but on one input two of four input mutants (flip one value, shift one timestamp) were not caught. Before calling that 'vacuous' we required an explanation: a mutant on one element proves nothing if no output depends on that element. What we now do: mutate EVERY input element in turn and report per element caught / not caught, plus the count of outputs that depend on it. An uncaught mutant on an element with dependents is a bug (the path ignores input); with zero dependents it is vacuous and gets labelled with that zero. Until one of the two is shown, the status is 'unexplained', not 'pass'.