I feel you. Been in this data-analysis space for years and have talked to most vendors, I’ll stay objective.

It’s tough science (geometry + geospatial). Points vs polygons vs non-uniformed polygons, unstable yield streams, etc. (to be fair, John Deere’s pipeline is one of the better ones).

From what you shared, the block areas differ between platforms.

• A common issue is projection math. You shouldn’t compute areas in global/web projections, but most yield files come that way. The farther from the equator, the bigger the distortion (Canada/South Africa see this amplified).

• Next level: the land isn’t flat. True area ideally accounts for topography/earth geoid. AFAIK no one calculates this in precisionAg.

• Any harvesting settings in granular applied, like flow delay, cut width, overlap handling, and auto-header width?

If you want, upload the data and I’ll run a neutral cross-check in our system (under free trial):

• Run trials in our system. You have an option to use raw and interpolated data (this could be also part of the error). 

• You can make quite advanced trial which compares not static areas, but for example rate1 vs rate2.

• Build a 3D field surface, overlay the trials on it

• Compare against historic vigor/soil/topography , helps catch cases where trials accidentally compare non-comparable areas (not saying that’s yours, but it happens)

• Checked as-planted data? Usually there is a shift between Rx & as-applied, and more accurate to compare trials based on as-applied/planted. How to measure Rx vs As-Planted error rate.