Introduction

Stochastic mine planning acknowledges that ore grade, geometry, and geotechnical parameters are uncertain. Rather than optimizing against a single “best estimate” block model, stochastic workflows run the optimizer across multiple equally-probable realizations and aggregate the results. Two broad families of approach have emerged: non-consensus workflows, which treat each realization independently and summarize the distribution of outcomes, and consensus workflows, which first identify a shared design footprint that performs acceptably across all realizations and then optimize within it. Each approach reflects a fundamentally different philosophy about how uncertainty should influence mine design.

Non-Consensus Stochastic Workflows: How They Work

In a non-consensus workflow, the optimizer — whether a stoping limit optimizer, a cave footprint finder, a minable shape optimizer, or a scheduling engine — runs independently on each realization. Each run produces its own stoping limit, caving footprint, mine design, and reserves estimate. The results are then aggregated statistically, typically reporting P10, P50, and P90 outcomes for NPV, tonnage, grade, and other key metrics.

Advantages

Full preservation of uncertainty information. Because each realization is optimized independently, the full distribution of possible outcomes is preserved. The P10 result genuinely represents a pessimistic scenario optimized for that pessimistic geology, and the P90 result represents an optimistic scenario optimized for optimistic geology. This gives planners an honest picture of upside and downside.

Realization-specific optimality. Each design is locally optimal for its geology. If realization 3 has a particularly high-grade pod in the northeast, the optimizer for that realization will exploit it. The resulting P90 NPV is not artificially constrained by what is achievable under pessimistic geology.

Simplicity of implementation. The workflow is embarrassingly parallel. Each realization runs through the same deterministic pipeline independently. There is no inter-realization communication, no consensus-finding step, and no additional algorithmic complexity beyond the base optimizer.

Sensitivity analysis. By inspecting which design decisions change most across realizations, planners can identify the geological parameters that most influence the outcome — a natural sensitivity analysis that emerges from the workflow without extra effort.

Disadvantages

No single actionable design. A mine can only be built once. If 10 realizations produce 10 different stoping limits with different shapes, orientations, and extents, the planner is left with a distribution of designs rather than a design. Translating that distribution into a capital commitment requires judgment that the workflow itself does not provide.

Optimistic bias in aggregated metrics. The P90 NPV from a non-consensus workflow is the 90th percentile of “best possible NPV given that geology.” In reality, the mine will be built to a fixed design, and that fixed design will underperform its own-realization optimum on every realization except the one it was designed for. Aggregating realization-optimal results overstates achievable value.

Design comparisons are apples-to-oranges. When realization 1 produces a 4,300-block stoping limit and realization 7 produces a 3,600-block stoping limit with a different shape, comparing their NPVs conflates geological uncertainty with design uncertainty. It is hard to isolate the effect of grade variability from the effect of choosing different footprints.

Consensus Stochastic Workflows: How They Work

In a consensus workflow, a preliminary voting phase determines which blocks appear in the optimal design across a sufficient fraction of realizations. Blocks that are “hot” in at least a threshold proportion of realizations — say 50% or 67% — form the consensus footprint. A single stoping limit mesh is then constructed from this consensus block set. In Phase 2, the optimizer runs once against this consensus mesh to produce a single mine design. In Phase 3, reserves are evaluated by running each realization’s grades through that fixed design, producing the P10/P50/P90 distribution for a design that could actually be built.

Advantages

A single actionable design. The consensus workflow produces one stoping limit, one mine design, one schedule. This is what engineers, financiers, and regulators need. The uncertainty is quantified around a fixed design rather than across a family of different designs.

Honest uncertainty quantification. Because Phase 3 evaluates all realizations against the same fixed design, the P10/P50/P90 distribution reflects genuine grade uncertainty for a design that will actually be executed. There is no optimistic bias from realization-specific optimization.

Conservative capital commitment. Blocks that only appear in high-grade realizations are excluded from the consensus footprint. The resulting design is sized to what can be justified under median or conservative geology. This naturally guards against over-commitment of capital to infrastructure that will only pay off under optimistic assumptions.

Comparability of reserve estimates. Since all realizations are evaluated against the same design geometry, differences in reported NPV across realizations are purely attributable to grade variability. The design uncertainty component is eliminated, making the sensitivity analysis cleaner.

Alignment with regulatory practice. Reserve reporting standards generally require a single mineral resource and reserve estimate associated with a defined mine plan. A consensus workflow naturally produces this, while a non-consensus workflow requires an additional step to select or construct a reportable design from the distribution.

Disadvantages

Value destruction by construction. The consensus footprint is by definition smaller than the optimal footprint for any individual realization. High-grade blocks present in only some realizations are excluded even when they would be highly profitable in those realizations. The P90 NPV of a consensus workflow will always be lower than the P90 NPV of a non-consensus workflow, because the consensus design cannot exploit realization-specific upside.

Threshold sensitivity. The shape and size of the consensus footprint depends strongly on the voting threshold. A 50% threshold produces a larger footprint than a 67% threshold, potentially including blocks with significant geological uncertainty. Choosing the threshold is itself a risk management decision that requires engineering judgment, and the workflow provides no automatic guidance on what threshold is appropriate.

Loss of realization-specific structure. The consensus footprint is the intersection of multiple stoping limits, each of which is a minimum-stope-width connected solid. Depending on how variable the realizations are, the intersection may be considerably smaller and differently shaped than any individual result, potentially sacrificing the minimum-stope-width connectivity that makes each individual result physically valid. Careful implementation is required to ensure the consensus mesh remains a valid, constructible solid.

Computational overhead. The consensus workflow requires running the optimizer once per realization in Phase 1 (same as non-consensus), plus an additional mesh construction and consensus-finding step, plus a second round of reserve calculations in Phase 3. For large deposits with many realizations, this is materially more expensive than a single-pass non-consensus workflow.

Sensitivity to reference model choice. Many consensus implementations use a reference block model to define the grid and extent within which voting occurs. If the reference model is unrepresentative of the realization ensemble — for example, if it is the highest-grade realization — the consensus footprint may be biased toward that model’s geometry.

Comparative Summary

Number of designs produced — non-consensus generates one optimized design per realization. Consensus produces a single shared design for the entire ensemble.

Actionability — non-consensus requires a post-processing step to select or construct a buildable design from the distribution. Consensus is directly actionable.

NPV distribution — non-consensus distributions are optimistically biased because each realization is evaluated against its own best-case design. Consensus distributions are honest — all realizations are evaluated against the same fixed design.

Capital risk management — non-consensus manages risk implicitly, typically by selecting the P10 design for execution. Consensus manages risk explicitly through the voting threshold, which directly controls how conservative the footprint is.

Sensitivity analysis — non-consensus conflates grade uncertainty with design uncertainty, making it harder to isolate the effect of geology alone. Consensus eliminates the design variable, so differences in outcome across realizations reflect grade variability only.

Regulatory reporting — non-consensus requires an additional step to identify a single reportable mine plan. Consensus is directly compatible with reserve reporting standards.

Computational cost — non-consensus is cheaper, requiring one optimizer pass per realization. Consensus requires the same Phase 1 passes plus a consensus-finding step and a second round of reserve evaluation.

Upside capture — non-consensus captures full realization-specific upside. Consensus caps return at what the consensus footprint can deliver, trading upside for predictability.

When to Use Each

Non-consensus workflows are well suited to early-stage studies where the primary goal is understanding the range of outcomes rather than committing to a design. They are also appropriate when the realizations are geologically similar enough that the distribution of optimal designs is tight — in that case, any reasonable design lies close to all realization optima and the distinction between consensus and non-consensus matters less.

Consensus workflows are appropriate when a capital commitment must be justified, when the realizations are sufficiently variable that non-consensus designs differ materially, or when regulatory reserve reporting requires a single defined mine plan. They are also the right choice when the risk philosophy of the project favors conservative capital sizing — accepting lower expected value in exchange for reduced probability of a significantly loss-making outcome.

In practice, a sophisticated planning workflow should use both: a non-consensus pass to understand the geological uncertainty landscape, followed by a consensus pass to produce the reportable design. The gap between the two P90 NPV figures is itself informative — it quantifies the value of flexibility, the cost of committing to a single design under uncertainty, and the potential upside if the geology proves better than the median.

Availability in GeoMine

Both workflow families are available in GeoMine as plugins or through its interactive automation scripting interface.  For further information, please Contact Us.