Resource estimation at gold deposits is frequently constrained by geological models built upon standardized deposit-type assumptions: rigid lithological and alteration frameworks that prioritize global consistency over fidelity to the underlying mineralization system. The Hemlo Gold Mine has utilized numerous genetic interpretations throughout its life to define and constrain mineralization domains. While practical in their original contexts, these boundaries introduce persistent statistical issues when applied at deposit scale, compromising statistical fidelity to ensure adherence to deposit classification frameworks.
A recent series of targeted updates challenged this convention, demonstrating that dynamically defined system boundaries, rooted in core geological principles, produce more representative and statistically defensible models than those derived from prescriptive deposit archetypes. This reassessment was grounded in deposit-scale petrogenetic analysis, initiated through significant collaboration with underground geologists whose first-hand structural and lithological observations provided the foundation for reinterpreting the architecture of the deposit’s mineralized domains.
Hemlo exhibits characteristics of orogenic, epithermal, and porphyry mineralization styles, each operating under distinct P-T regimes and focusing mineralization through fundamentally different trapping mechanisms. These trap types are spatially superimposed but mechanistically and geochemically discrete, meaning each defines a volumetrically different region capable of hosting mineralization under different local controls. A single prescriptive classification cannot adequately represent this complexity, and imposing one introduces systematic error into any domain framework built upon it. Mapping the distribution of these trap types provided the first-order framework for constraining mineralization potential, which was subsequently used as the foundation for domain refinement using geochemical assemblages and structural interpretations.
This approach allowed populations of similar mineralization style, including disseminated and vein-hosted gold, to remain internally coherent across boundaries that would previously have separated them on lithological grounds. Conversely, geologically distinct populations previously grouped under a shared lithological domain were separated based on contrasting event signatures. This has the effect of creating stationarity within the dataset through the isolation of mineralization style, producing populations representative of discrete mineralizing events.
Further refinement was completed using a multi-evidence framework integrating updated geological logging, revised structural interpretations, new alteration models, and geochemical assemblages diagnostic of the identified mineralizing events. Physical domain boundaries derived from this framework were used as the basis for Indicator RBF grade shell generation.
The statistical results were significant. Mean external grade variance decreased by 46% while internal mean grade increased by 15%, reflecting improved coherence of the data populations. Internal dilution decreased by 10%, while samples over cutoff outside of the model were reduced by 23%, indicating higher model accuracy. Internal and external CoV were reduced by an average of 10%, indicating more consistent data populations despite removal of lithological boundaries. Together, these outcomes represent a meaningful reduction in conditional bias affecting previous model iterations.
Hemlo demonstrates that where domain frameworks drift from underlying geological controls, incremental refinement is unlikely to resolve the core problem. Returning to the fundamental question of what drives grade distribution, and building the model architecture around that answer, produces both statistically and geologically superior results.