178 At Targets 6 and 16, the deficiencies of the historical drilling programs were geometric rather than related to absence of mineralization. At Target 16, historical drilling had intersected the zone margin but failed to follow up with geometrically appropriate hole orientations. Fan-pattern drilling and orientations perpendicular to mineralization resulted in high-grade intercepts being distributed across long composite intervals, effectively obscuring the grade signal. The DL model’s spatial representation framework preserves high-grade information in a manner structurally different from Kriging compositing, which is the mechanism by which these zones were recovered as drill targets. Table 3 – Drilling efficiency comparison, Chilean site Metric Site Baseline (2020–2021) AI-Guided Program Cost per kT Cu verified Baseline 25% lower Total meters drilled Comparable 2,200 m Holes intersecting HG Variable 11 of 11 Cu added to Measured — 7.7 kT (~USD 64M insitu) Figure 8 – Pre-drill estimated values and post-drill verified results for each target zone 4.2. Western Australian Orogenic Lode Gold The Australian study deposit is an underground orogenic lode gold system characterized by nuggety grade distributions and quartz-vein-hosted mineralization. Mineralized zone (MZ) logging is employed extensively for domain definition; however, MZ codes are applied across multiple pre-mineral lithological units, and both barren and bonanza-grade assays occur within short spatial intervals. Consequently, MZ classification constitutes an insufficiently reliable sole basis for grade estimation. Application of the PSA identified specific lithological and alteration codes that correlated more reliably with gold grade than MZ logging alone. When incorporated as pathfinder channels, the DL model recovered mineralized intervals that the site’s Cokriging (CIK)-based model had not identified. The improvement was most pronounced in deposit sectors where MZ logging had been applied inconsistently by different geologists over successive campaigns—zones in which the Kriging model propagated logging inconsistency as spatial noise, while the DL model, operating on PSA-screened inputs, resolved through it. Drill targeting based on the model confirmed economic gold grades in volumes the site model had
RkJQdWJsaXNoZXIy MTM0Mzk2