179 classified as waste, validating the pathfinder approach in a structurally complex, nugget-effect-dominated system in which conventional domain-based estimation exhibits documented performance limitations (Glacken and Blackney, 2022; Sims, 2023). 4.3. Kazakh Orogenic Lode Gold The Kazakh study site is an orogenic lode gold deposit in western Kazakhstan, presenting the most challenging data environment of the three sites investigated. The logging database had been compiled over a more extended operational history with greater variation in geological conventions and code usage. Application of the PSA identified a smaller set of codes meeting the significance threshold than at either the Australian or Chilean deposits—a direct quantitative reflection of the reduced signal-to-noise ratio in the logging data. The DL model trained on PSA-screened inputs nonetheless outperformed the site conventional estimation in prospective blind reconciliation, though the magnitude of improvement was smaller than at the Chilean site. This outcome is consistent with the lower signal content of the available logging data. The result demonstrates both that the methodology adds value under conditions of degraded data quality and that the magnitude of that improvement scales predictably with the information content of the input data. The PSA adapts to this constraint rather than imposing uninformative inputs on the model. 4.4. Summary Across Deposits Table 4 – Cross-deposit comparison Deposit Type Commodity DL vs. Kriging (Recall) Drill Outcome Chile (Candelaria–PdC) IOCG, manto Cu 1.67× improvement 7.7 kT Cu verified; 11/11 holes hit HG Western Australia Orogenic lode Au 6x improvement, reconciled in drilling between 2023-2024 Improved, strongest in inconsistent MZ zones Economic Au confirmed in wasteclassified zones Kazakhstan Orogenic lode Au 4x improvement 35koz Au verified; 11/13 holes hit HG The pattern across sites is consistent: DL resource estimation incorporating PSA-screened geological logging outperforms conventional Kriging at all three study sites. The magnitude of outperformance is positively correlated with geological logging data quality. The methodology did not require fundamental architectural modifications between deposit types—the same framework, screening approach, and evaluation protocol was applied in each case, with the PSA adapting input selection to the signal characteristics of local data.
RkJQdWJsaXNoZXIy MTM0Mzk2