Override Value Index - Geology (OVI-G)

GMI-MIN-008MiningIntegration-Required

Definition

Economic value gained or lost from geologist overrides of AI grade predictions, measured by comparing override outcomes to what would have occurred had the AI prediction been accepted.

The Measurement Gap

AI orebody models produce grade predictions that geologists review before operational use. In practice, experienced geologists sometimes override the model's prediction based on local knowledge, visual observations, or professional judgment about geological features the model may not capture. Traditional metrics have no mechanism to evaluate whether these overrides improve or degrade prediction quality. Standard reconciliation reports show overall model accuracy, which blends the performance of accepted and overridden predictions into a single number.

Legacy metrics like ore recovery ratio, model bias, mean absolute error, and cost per ton cannot distinguish between model performance when left alone and model performance after human modification. Physical Availability, MTBF, MTTR, and PM Compliance are equipment metrics entirely disconnected from geological prediction accuracy. Even advanced geostatistical validation metrics test the model's structural properties, not the combined human-model system performance. Without measuring override value, organizations risk two opposite failures: suppressing valuable expert judgment in favor of blind model trust, or allowing overrides that systematically degrade prediction quality because the geologist's mental model is outdated.

Decision Supported:

Are geologist overrides making grade control better or worse, and should the team trust the model more, trust the geologist more, or improve the feedback loop between them?

Formula

Plain Language: Separate all grade predictions into two groups: those the geologist accepted as-is and those the geologist modified. For each group, calculate the average absolute error between the prediction and the actual post-extraction assay result. OVI-G is the ratio of the overridden group's error to the accepted group's error. If overrides have lower error than acceptances, the ratio is below 1.0 (overrides are improving quality). If overrides have higher error, the ratio exceeds 1.0 (overrides are degrading quality).

Notation:

OVI-G = Accuracy_overridden / Accuracy_accepted

Where Accuracy = mean |actual - prediction| averaged across instances. (Lower error = higher accuracy, so OVI-G < 1.0 means overrides improve accuracy.)

Variable Definition Unit
Accuracy_overridden Mean absolute prediction error for predictions where the geologist modified the model output Grade units (% Cu, g/t Au)
Accuracy_accepted Mean absolute prediction error for predictions where the geologist accepted the model output unchanged Grade units
actual Post-extraction assay result for the predicted zone Grade units

A result below 1.0 means overrides improve accuracy on average. Above 1.0 means overrides degrade accuracy. A result of 1.0 means overrides make no difference.

Proposed formula, pending validation pilots.

At a Glance

Unit
Ratio
Trend Direction
Higher; Increasing above 1.0
Reporting Frequency
Monthly
Category
Predictive Geology Model Integrity
← Model Boundary Attribution Rate (MBAR)Lifecycle Model Survivorship Bias Index (LSBI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment