Soma Gold is installing an XRT sensor-based sorter at El Bagre in Colombia after positive test work at TOMRA’s facility. The goal is straightforward: reject waste early, stabilize plant feed, and lower unit costs. For a small producer handling both in-house and third-party material, this is a practical lever to defend margins without adding new ounces in the ground. The upside is real if recoveries hold and mass rejection is meaningful. The risk is equally clear if mineralization proves harder to see than the test work implies.
El Bagre processes mixed sources with variable grades and textures, which drives volatility through crushing and grinding. X-ray transmission sorting tackles this by separating particles based on atomic density before energy-intensive milling. When it works, mills see a more consistent, higher-grade feed. TOMRA’s COM Tertiary XRT 1200, combined with its OBTAIN and CONTAIN machine-learning packages and a high-speed ejection module, is designed to keep precision high even under full belts and small size fractions. The reported bench testing showed consistent upgrade factors across size ranges, and the air ejection module can cut compressed air use by up to 70 percent, reducing operating cost. Integration into the main flowsheet, rather than a side-stream, signals Soma is going for a plant-wide effect on throughput and stability, not a boutique upgrade.
The physics matters. XRT excels when density contrast between valuable and waste phases is strong at the particle scale. Native gold is very dense, but in many orogenic and vein systems it occurs as fine inclusions, locked within sulfides, or as micron-scale gold refractory in arsenopyrite. Pure quartz with visible gold is ideal for XRT; disseminated fine gold in quartz-sulfide mixtures is harder. TOMRA’s OBTAIN and CONTAIN models aim to flag inclusion-type mineralization patterns that older algorithms missed, but the sorter still needs particles large enough and with sufficient density contrast to be detected. Performance depends on fragmentation, moisture, belt presentation, and the prevalence of barren look-alikes. Test centers often show best-case results under controlled conditions. At site, humidity, dust, worn screens, or a shift in ore texture can degrade signal quality and confuse the model. Expect a learning curve and ongoing calibration.
Sorting only creates value if the plant discards a meaningful fraction of tonnage while retaining most of the gold. A simple yardstick: if you reject 30 to 50 percent of feed mass while keeping 85 to 95 percent of contained metal, head grade rises 1.5 to 2 times, unlocking mill capacity and energy savings. That can translate into higher production or lower all-in sustaining costs, depending on where the bottleneck sits. Energy is a clear lever; grinding is power-hungry, so every tonne of waste diverted reduces cost. Consumables wear also falls with lower hard-rock throughput. For El Bagre’s mixed feed, a sorter can also tighten reconciliation and reduce dilution from third-party ore. Watch the trade-offs: higher precision usually means lower throughput or tighter size windows. If the sorter under-recovers, gold goes to the reject belt, and the upgrade math collapses. The payback runs through three numbers that Soma should publish post-commissioning: mass pull to accept, gold recovery to accept, and mill throughput change.
Pre-concentration adds a new control point. To realize the upgrade, crushing and screening must deliver a stable particle-size distribution, with fines either bypassed or handled by a dedicated sorting setting. High belt occupancy can compress residence time and reduce detection quality; the hardware Soma selected is built for dense flows, but operators still need to manage feed rate and bed depth. Moisture control is not optional; wet fines can mask density signals and increase misclassification. Regular assay audits of accept and reject streams are critical to manage metal loss risk, especially in the first months. The promise of AI-enabled recognition is only as good as the training data; models need updating as the orebody evolves. In Colombia’s humid tropics, uptime depends on robust dust control, vibration monitoring, and prompt parts availability. Working with a local integrator like DISMET should help, but investors should assume ramp-up variability.
Sorting can be a powerful tool for operations that buy ore from small-scale miners. It can standardize feed grade and remove waste rock that would otherwise consume plant time. That said, commercial terms get trickier when a sorter stands between the weighbridge and the mill. Expect more attention on sampling protocols and settleable grade determination. If suppliers see higher reject rates, they may press for different pricing or seek alternative buyers. Soma will need tight chain-of-custody procedures and transparent reconciliation to avoid disputes while capturing the efficiency gains. The company also must watch for contamination from artisanal sources that introduce deleterious elements, which can skew sorter performance and downstream metallurgy if not screened.
Juniors across the Americas are delaying fieldwork due to shortages of geologists and drill crews, pushing service costs higher and elongating schedules. Against that backdrop, investing in process technology that lowers operating cost per tonne can be rational. It reduces dependency on incremental headcount and can defer larger capex on grinding. But complexity adds its own staffing needs: sorter operators, maintenance techs, and data analysts to manage model drift. If labor remains tight, training and retention become part of the execution risk. The choice to integrate into the main flowsheet suggests Soma expects a full-plant benefit; it also raises the stakes if the unit underperforms, since bypass circuits then carry the load.
Mining equities outperformed broader markets into year-end as gold strengthened and juniors with advancing projects led gains. But the market is still punishing overpromises, especially on early-stage data that does not translate into mine cash flow. Technology that demonstrates direct cost and throughput gains tends to get better reception than speculative step-outs. Strategic partnerships and project-level financings are letting companies stretch dollars, and vendors increasingly co-develop solutions. TOMRA’s local integration and the test-center-to-plant path fit that pattern. For Soma, this is not a discovery catalyst; it is an operating-margin catalyst. Rerating potential will come only after credible operating data are in hand at commercial scale.
– Verified mass rejection and gold recovery to accept over at least a 30-day continuous run
– Assay audits of the reject stream to quantify gold losses, including spot checks of fines
– Ability to maintain precision at high belt occupancy and across size fractions in wet season conditions
– Reconciliation outcomes on third-party ore and any changes in procurement terms
Bottom line: sensor-based sorting could be a sensible, near-term margin tool for El Bagre, if site performance matches test work. The geology will set the ceiling for what XRT can accomplish. Execution will determine how close Soma gets to it.
