Test the same accession twice in one year — different substrate, different dormancy treatment — and you'll get two numbers. Before any analysis, they must become one. The choice of how encodes a belief about what the replicates measure.
Taking the maximum says: treatments differ in their ability to wake seeds up, and the best treatment is the truest reading of how many are actually alive. A 20% and a 60% in the same year means the first treatment failed the seeds, not that the seeds failed. Pooling the counts says: the replicates are exchangeable draws from one lot, so combine the evidence — statistically cleaner, but it averages a failed dormancy treatment into the viability estimate.
For wild species with stubborn, poorly understood dormancy, max is usually the defensible default. For crop seeds with standardized protocols, pooling is. Either way: one strategy, declared, applied everywhere.