Where a gemstone or a grain of gold comes from is no longer a question only paper trails can answer. Emeralds show what is possible: using the largest global dataset available, more than 1,000 LA-ICP-MS analyses of gem-quality emeralds, machine learning models trained on just eight trace elements (Li, V, Cr, Fe, Sc, Ga, Rb, Cs) determine geographic origin with 96.8% accuracy. Gold is the harder problem, and the more urgent: illegal mining in the Amazon drives mercury contamination and opaque supply chains. We are now applying the same architecture to native gold, building the curated reference datasets needed to test whether source-related groupings can be resolved across the geologically heterogeneous Amazonian craton, a stringent challenge given low trace-element concentrations and overlapping geochemical signatures. Built on well-documented reference collections such as Harvard's and partner Brazilian institutions, this work provides an initial basis for geochemical screening of mercury-linked illicit supply chains, and a template for the future of mineral traceability.