You're totally right. In a few words - you can't trust any software until it's open sourced and you can in detail check how algorithms work on their github.

In our case - we don't make it open source (just not enough developers, only competitors will take it for free), but try to explicitly via tutorials and in-app hints communicate how it works. 

An example - half of the area doesn't have yield monitor data - do you want to count it as 0 VS avg of other regions VS simple interpolation VS prediction synthetic yield? All 4 options will lead to different results.

Different software package do it in their own 'black box' way.

IMO as a precisionAg software vendor we need to provide: 1) Raw data visualizations; 2) Processed data based on client's choice (for example fill gaps with synthetic yield prediction). Also giving too many options not allow to create simple user interface.