Genetic Gain per Year

A breeding programme is judged on progress per year, not progress per cycle, and the two can point in opposite directions. A programme that achieves a large response over an eight-year cycle is beaten by one achieving half as much every three years.

Written out fully, annual gain is $\Delta G = i h \sigma_A / L$ — selection intensity, the square root of heritability, additive standard deviation, all divided by cycle length. Three of those four terms are difficult and expensive to improve. Intensity runs into diminishing returns and inbreeding. Heritability improves only with more replication and more locations, which costs money linearly. Additive variance is a property of the germplasm and shrinks under the very selection you are applying.

Cycle length is the exception, and it is why the modern breeding literature is dominated by it. Halving $L$ doubles annual gain outright with no diminishing return. Off-season nurseries in the opposite hemisphere, speed breeding under extended photoperiods, doubled haploids that collapse six selfing generations into one, and genomic selection — which predicts merit from markers before a plant has ever been yield-tested — are all attacks on the denominator rather than the numerator.

Measure the cycle honestly. It runs from making a cross to using its selected descendants as parents of the next cross, which includes every generation of line fixing and every year of yield testing. Programmes that quote only the yield-testing years understate $L$ badly and overstate their own rate of progress.