175 Figure 2 – Geological cross-section of the Chilean IOCG study site showing the host stratigraphy, fault-controlled mineralization, and the relationship between Manto-style and feeder-structure orebodies. Prospective blind reconciliation spanned eight consecutive quarters (2021–2022), with all models trained exclusively on pre-2021 data. Results at the 0.5% Cu cut-off are presented in Table 1. Table 1 – Chilean IOCG forward-facing reconciliation (2021–2022) Model Missed Mineralization Reconciled HG vs. Kriging Kriging (site model) 83.2% 16.8% Baseline Method 1: D(Cu) ~ D(Cu) 82.8% 17.2% ~1× Method 3: D(Cu, ZFCU) ~ D(Cu, ZFCU) 71.8% 28.2% 1.67× Figure 3 –False positive mineralization rates (>0.5% Cu) and (b) missed mineralization rates (>0.5% Cu) for 2021–2022 across model configurations and the site Kriging model. Method 1—DL trained on assay data only—produced performance virtually indistinguishable from that of the site Kriging model, confirming that the DL architecture itself does not confer performance gains
RkJQdWJsaXNoZXIy MTM0Mzk2