I will try to answer this using my experience with the soybeans and using two other examples involving field pea and lentil grown in Saskatchewan, Canada.
For grain legumes, harvest index (HI) is calculated as the ratio of harvested grain dry mass to the total above ground plant dry mass.
HI = grain dry mass/total above ground plant dry mass
It is important to understand that HI can vary widely, depending on the timing of sampling and the extent of leaf fall before the full maturity. For crops which drop leaves before reaching the full maturity (as in the case of indeterminate grain legume varieties), HI should be calculated using the maximum biomass dry matter, including the fallen leaves. Not just the standing above ground dry matter at harvest.
To illustrate this point, I included three examples in the attached Table. All examples are from indeterminate grain legumes grown in Canada. The example for soybean is from one of our experiments at the University of Guelph and the two other are from Malhi et al. (2004). In all cases, sequential plant sampling has been carried out in regular intervals (1 to 2 weekly) during the life span of the crop.
You can see from the three examples that HI could be considerably overestimated if one fails to account for the dry mass of fallen leaves before the full maturity. Your question is very pertinent. There is a need to obtain more reliable field HI data, especially with the increasing importance in field level C accounting (in the face of global change).