geological team drilled the zone, confirmed economic grade, and revised their structural model. The model showed an inconsistent structural interpretation that was not fully supported by the existing drilling data. This represents, in the authors’ assessment, an appropriate operational relationship between computational modelling and geological domain expertise in mining contexts: the model characterizes what the data indicate; geologists determine what it signifies. The historical drilling deficiencies at Targets 6 and 16 also merit methodological discussion. In both cases, mineralization was not absent from the historical drill record—at Target 16, a historical hole had intersected the zone boundary. The failures were geometric: hole orientations perpendicular to ore shoot plunge, and fan-pattern designs that precluded spatially targeted follow-up. High-grade intercepts composited across elongated intervals became indistinguishable from subeconomic material in the Kriging model. The DL model’s representation of grade information within its spatial learning framework—as distinct from the averaging mechanism of compositing—constitutes a material practical advantage for structurally complex deposits such as IOCG systems. 5.4 Limitations The methodology is dependent on geological logging quality. Deposits in which logging is inconsistent, poorly standardized, or dominated by codes with limited discriminating power will yield smaller improvements relative to conventional estimation—a conclusion directly demonstrated by the Kazakh case. The PSA mitigates this dependency by filtering low-information inputs, but cannot manufacture predictive signals from data that does not contain exist. Prospective validation was conducted at all three sites; however, the depth of validation differs. The Chilean site provides the most comprehensive dataset: eight quarters of blind reconciliation and a completed 2,200 m drill program. The Australian and Kazakh sites have shorter validation windows. Continued operational deployment will generate additional data against which long-term performance stability can be assessed. An organizational dimension not captured by technical performance metrics also warrants acknowledgement. Mine planning workflows are institutionally structured around Kriging-based outputs, and the integration of DL model predictions requires adaptations that extend beyond software installation. Geological and engineering teams must exhibit a dependence on the in model outputs, a process that is built incrementally through demonstrated field performance rather than through hypothetical exercises. The Chilean drill program provides a substantive evidentiary foundation, but each new operational site required its own period of parallel evaluation before model outputs were incorporated into binding mine plan decisions. 5.5 Methodological Transferability The methodology was designed to be deposit-agnostic with respect to commodity, geological setting, and data architecture. The PSA, data integration strategies, confidence quantification framework, and evaluation protocol make no assumptions specific to any particular
RkJQdWJsaXNoZXIy MTM0Mzk2