Arbor 
Trace the source of every truckload
Understand where the ore came from
From model and blast block to bucket and truckload
Build an accountable chain of ore movement
Mining solution · Grade Control Copilot
Every bucket, every truckload
Directed to a better destination
AI-native ore grade management
Reduce dilution and ore loss, stabilise grade
Highlights
Bring the grade model into operations
Evidence for every step
Trace the source of every truckload
Assess confidence first
Then decide whether to test and where to send it
Turn grade uncertainty and testing costs
Into explainable routing recommendations
Connect every optimisation
To actual operational accounts
Calibrate predictions with assays; reconcile execution with production records
Account for benefits and additional costs together
01 / 03
Ore-flow operations graph · OreFlow Graph
From a block of ore
To the final assay
The model describes the material; operational records show where it went. Connect objects, events and results so every recommendation can be traced to evidence.
One truckload, one traceable chain of custodyLocations and connections illustrate the proposed mechanism
One model, with its version and confidence retained
Block grade, impurities and uncertainty provide the starting point. Blasting, loading, haulage, plant feed and assays preserve material identity along the way, allowing final results to be traced back to their sources.
Drones · Frequent geometry updates
As the site changes
Update the basis for mining
See muckpile geometry, mining-face advance and volume changes, and align the geological model with the actual site.
Bring the current site into the modelLayers illustrate the mechanism, not actual survey results
Drones update the geometry
Geology and assays establish grade
Align current terrain, model versions and blast-movement evidence in one coordinate system to review mining boundaries. Ordinary aerial imagery does not directly measure underground Fe grade; geometry updates also require registration-accuracy and error checks.
Excavators and haul trucks · Source estimation
The grade of this truckload
Starts with every loading event
Interactive mechanism demonstration · Not production dataOne truckload
One truckload
May contain material from different blocks
Associate bucket positions with blocks, then calculate the mass-weighted blend. Try changing the loading share from Area A.
Area A · Fe 62%Area B · Fe 54%
% FePredicted truckload mean. Source-matching error, geological uncertainty and measurement scale must also be considered.
Truckload grade = Σ (source mass fraction × source grade)
Trace each bucket where precise bucket positions are available
Keep uncertain sources unknown
Loading times, bucket trajectories, mining boundaries and weighing jointly determine confidence in material origins. Without bucket-level data, organise records by verifiable mining face or batch rather than forcing a precise source estimate.
Decisions under uncertainty
The same average grade
Different levels of confidence
A mean alone is not enough. Greater model uncertainty makes the probability of meeting the target grade more important.
Probability-model demonstration · Mean Fe 58.3%
Probability of reaching 57% Fe
%Clearer sources and stronger evidence narrow the prediction distribution. Continue calibrating it against actual assays.
Grade, source and scale
Together determine confidence
Together determine confidence
Thresholds and probabilities illustrate the method. Actual constraints also include impurities, recovery and processing requirements.
Expand to see how the model represents uncertainty
Truck mean μ = Σ wᵦμᵦ; variance Var(g) = wᵀΣw plus source matching error and scale error. The illustration uses a normal distribution with mean 58.3% and standard deviations of 0.65 and 1.8 percentage points respectively. Validate the distribution, correlations and parameters against actual mine data.
Selective measurement · EVSI
Use testing
Where it adds the most value
Would another measurement change the decision? Would the resulting value cover testing, waiting and diversion costs?
Not every truck needs a detour for testingSelect measurements according to decision value and site conditions
Truckload decision simulation · Amounts in CNY
Test first, then update the destination recommendation
Truckload evidenceSelective additional testingReassess
If an additional test is expected to improve decision value by CNY 900 and testing, waiting and diversion cost CNY 350, the net value of information is CNY 550. In this example, further testing may be worthwhile.
EVSI = expected best value after testing − current best value − additional cost. This example does not represent actual testing benefits. Match the sensing method to the mineral, particle size and sampling representativeness.
Specialist optimisation + Human confirmation
Grade, capacity and haul distance
Together determine the next destination
Recommendations must meet production constraints and be executable on site. AI organises evidence, specialist models solve the optimisation problem, and dispatchers and geologists confirm the result.
Simplified constraint demonstration · Not live dispatch
Recommended destination
Crusher
Grade constraints are met and the crusher has capacity in this example. The recommendation is to send the load directly to the plant under the approved plan, reducing additional handling.
Current haul truckCrusher
Recommendation → Geology / dispatch confirmation → Execution
Retain input versions, decision evidence, approvers and actual unloading points. Validate interface write-back and the scope of on-site control separately.
Quality constraints
Minimum Fe grade, maximum SiO₂, Al₂O₃ and P contents, and the probability of meeting requirements
Production constraints
Crusher throughput, stockpile capacity, material compatibility and the current plan
Economic constraints
Compare ore value per tonne with testing, diversion and rehandling costs
Assay reconciliation · ROI Ledger
Every optimisation
Reconciled with operational accounts
Use actual assays and production records to reconcile grade, tonnage and destination. Account consistently for recovered ore, diverted waste rock and additional costs.
Prediction → Execution → Assay → Reconciliation → CalibrationTrace final results back along the ore flow to support the next improvement cycle
Net value under these assumptions
¥100 tonnes × CNY 200 + 200 tonnes × CNY 80 − CNY 15,000
Illustrative assumptions: CNY 200 per dry tonne in net value from reducing ore loss and CNY 80 per dry tonne saved by diverting waste rock. The mine must confirm actual values; do not count the same benefit twice.
This demonstrates an accounting method, not a project-return forecast. Establish a baseline using comparable mining areas and shifts, controlling for prices, moisture, inventory and other technical changes.
Less dilution
Verify the tonnes of waste rock entering the ore stream and identify where misclassification occurred
Less ore loss
Verify the tonnes of ore sent to waste dumps and retain material awaiting testing or a decision
Stable grade
Compare plant-feed assays with prediction distributions to assess whether bias and variability are decreasing
Expand the accounting method and evidence boundaries
Agree the denominators, material classifications and treatment of unresolved inventory before measuring dilution and ore loss. Reconcile dry-tonne inputs, outputs and inventory changes. Validate effects through historical reconstruction, matched comparisons and controlled pilots. Improvement percentages from research illustrations are not presented as Arbor results.
Start with one mining area
Prove one complete operational loop
In practice
Connect existing geological models, FMS, drones and assay systems. Start with verifiable data and test recommendations against real production.
01 · Historical reconstruction
Reconcile the past firstAlign models, coordinates, truck trips and assays to reconstruct material flow in one mining area and identify missing source information and misclassification.
02 · Shadow validation
Compare recommendations with current operationsRun recommendations alongside current production without changing instructions. Record when they differ, why, and the subsequent results.
03 · Controlled pilot
Expand gradually on evidenceBegin a limited pilot after human confirmation, then expand based on validation of quality, costs and interfaces.
Solution capabilities, data sources and implementation notes
This page presents a grade-management solution and proposed operating mechanisms for open-pit mines, not a list of delivered functions. Probability models, sensing methods, data granularity, integration, permissions and write-back capabilities depend on project assessment and validation. AI scenes, layers and interactions explain the concept; they are not customer site records.
The method draws on the supplied Grade Control Copilot strategy and ore-flow operations-graph research. Technical references include Maptek Grade Control、CSIRO mineral analysisand Palantir object and action modelling. References do not imply a partnership, complete interface compatibility or existing delivery by Arbor.