Definition
The share of a recommendation engine's error that traces to incomplete or incorrect farmer-elicited inputs (hybrid, planting date, tillage, manure, irrigation history) rather than to the engine's own inference, computed on units where the elicited input can be independently reconstructed from machine data or records.
The Measurement Gap
There is no metric that tells you how much of a recommendation's error traces to incomplete or wrong farmer-elicited inputs versus the model's own inference, which means input quality and model quality cannot be improved on purpose because they are measured as one.
Formula
EAES-AGR = (E_as_entered − E_corrected_inputs) / E_as_entered, where E is the engine's error against outcome on the reconstructable units, pooled per engine and per input field
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Whether to invest in input capture (machine-derived fields, defaults, validation prompts) or in the model, since the two are currently blamed for each other's errors.