minimum
The share of attempted replicates that must be usable before a verdict other than INCONCLUSIVE is given.
A half is a judgement rather than a measurement, and it is deliberately blunt. The concern it encodes is real: replicates fail because a sample was hard to fit, so the survivors are systematically the easier ones and the null built from them sits too low. A null that sits too low rejects too readily. Whether a half is the right line is a question for the validation pass rather than something to settle by argument.